From 673d24666331464de43a86f560992f730a1cb7d8 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Tue, 10 Feb 2026 16:42:04 -0700
Subject: [PATCH 01/27] :sparkles: Start processing tool responses with nemo
 check

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .../examples/nemocheck/k8deploy/deploy.yaml   |  14 +-
 .../examples/nemocheck/nemocheck/plugin.py    | 142 ++++++++++++++++--
 src/server.py                                 | 105 ++++++++++---
 3 files changed, 218 insertions(+), 43 deletions(-)

diff --git a/plugins/examples/nemocheck/k8deploy/deploy.yaml b/plugins/examples/nemocheck/k8deploy/deploy.yaml
index 5354549..0971cb8 100644
--- a/plugins/examples/nemocheck/k8deploy/deploy.yaml
+++ b/plugins/examples/nemocheck/k8deploy/deploy.yaml
@@ -34,22 +34,22 @@ spec:
             - name: MODEL_NAME # Currently only for logging.
               value: "meta-llama/llama-3-3-70b-instruct"
             - name: CHECK_ENDPOINT
-              value: "http://nemo-guardrails-service:8000/v1/guardrail/checks"
+              value: "http://nemo-guardrails-service:50053/v1/guardrail/checks"
             - name: PLUGINS_SERVER_HOST
               value: "0.0.0.0"
             - name: PLUGINS_ENABLED
               value: "false"
-            - name: PLUGINS_CLI_COMPLETION  
+            - name: PLUGINS_CLI_COMPLETION
               value: "false"
-            - name: PLUGINS_CLI_MARKUP_MODE  
+            - name: PLUGINS_CLI_MARKUP_MODE
               value: "rich"
-            - name: PLUGINS_CONFIG  
+            - name: PLUGINS_CONFIG
               value: "./resources/plugins/config.yaml"
-            # - name: CHUK_MCP_CONFIG_PATH  
+            # - name: CHUK_MCP_CONFIG_PATH
               # value: "./resources/runtime/config.yaml"
-            - name: MCP_SSL_ENABLED  
+            - name: MCP_SSL_ENABLED
               value: "false"
-            - name: MCP_SSL_CERT_REQS  
+            - name: MCP_SSL_CERT_REQS
               value: "0"
             - name: LOGLEVEL
               value: "DEBUG"
diff --git a/plugins/examples/nemocheck/nemocheck/plugin.py b/plugins/examples/nemocheck/nemocheck/plugin.py
index 82e96a7..4983795 100644
--- a/plugins/examples/nemocheck/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/nemocheck/plugin.py
@@ -36,8 +36,12 @@
 log_level = os.getenv("LOGLEVEL", "INFO").upper()
 logger.setLevel(log_level)
 
-MODEL_NAME = os.getenv("NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct")  # Currently only for logging.
-CHECK_ENDPOINT = os.getenv("CHECK_ENDPOINT", "http://nemo-guardrails-service:8000")
+MODEL_NAME = os.getenv(
+    "NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct"
+)  # Currently only for logging.
+CHECK_ENDPOINT = os.getenv(
+    "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000"
+)
 
 
 class NemoCheck(Plugin):
@@ -52,7 +56,9 @@ def __init__(self, config: PluginConfig):
         """
         super().__init__(config)
 
-    async def prompt_pre_fetch(self, payload: PromptPrehookPayload, context: PluginContext) -> PromptPrehookResult:
+    async def prompt_pre_fetch(
+        self, payload: PromptPrehookPayload, context: PluginContext
+    ) -> PromptPrehookResult:
         """The plugin hook run before a prompt is retrieved and rendered.
 
         Args:
@@ -65,7 +71,9 @@ async def prompt_pre_fetch(self, payload: PromptPrehookPayload, context: PluginC
 
         return PromptPrehookResult(continue_processing=True)
 
-    async def prompt_post_fetch(self, payload: PromptPosthookPayload, context: PluginContext) -> PromptPosthookResult:
+    async def prompt_post_fetch(
+        self, payload: PromptPosthookPayload, context: PluginContext
+    ) -> PromptPosthookResult:
         """Plugin hook run after a prompt is rendered.
 
         Args:
@@ -77,7 +85,9 @@ async def prompt_post_fetch(self, payload: PromptPosthookPayload, context: Plugi
         """
         return PromptPosthookResult(continue_processing=True)
 
-    async def tool_pre_invoke(self, payload: ToolPreInvokePayload, context: PluginContext) -> ToolPreInvokeResult:
+    async def tool_pre_invoke(
+        self, payload: ToolPreInvokePayload, context: PluginContext
+    ) -> ToolPreInvokeResult:
         """Plugin hook run before a tool is invoked.
 
         Args:
@@ -87,11 +97,11 @@ async def tool_pre_invoke(self, payload: ToolPreInvokePayload, context: PluginCo
         Returns:
             The result of the plugin's analysis, including whether the tool can proceed.
         """
-        logger.info("tool_pre_invoke....")
+        logger.info("[NemoCheck] Starting tool_pre_invoke")
         logger.info(payload)
         tool_name = payload.name  # ("tool_name", None)
         check_nemo_payload = {
-            "model": MODEL_NAME,
+            "model": MODEL_NAME,  # ideally optional
             "messages": [
                 {
                     "role": "assistant",
@@ -101,7 +111,9 @@ async def tool_pre_invoke(self, payload: ToolPreInvokePayload, context: PluginCo
                             "type": "function",
                             "function": {
                                 "name": tool_name,
-                                "arguments": payload.args.get("tool_args", None),
+                                "arguments": payload.args.get(
+                                    "tool_args", None
+                                ),
                             },
                         }
                     ],
@@ -109,31 +121,49 @@ async def tool_pre_invoke(self, payload: ToolPreInvokePayload, context: PluginCo
             ],
         }
         violation = None
-        response = requests.post(CHECK_ENDPOINT, headers=headers, json=check_nemo_payload)
+        response = requests.post(
+            CHECK_ENDPOINT, headers=headers, json=check_nemo_payload
+        )
         if response.status_code == 200:
             data = response.json()
             status = data.get("status", "blocked")
             logger.debug(f"rails reply:{data}")
             if status == "success":
                 metadata = data.get("rails_status")
-                result = ToolPreInvokeResult(continue_processing=True, metadata=metadata)
+                result = ToolPreInvokeResult(
+                    continue_processing=True, metadata=metadata
+                )
             else:
                 metadata = data.get("rails_status")
                 violation = PluginViolation(
-                    reason=f"Tool Check status:{status}", description="Rails check blocked request", code=f"checkserver_http_status_code:{response.status_code}", details=metadata
+                    reason=f"Tool Check status:{status}",
+                    description="Rails check blocked request",
+                    code=f"checkserver_http_status_code:{response.status_code}",
+                    details=metadata,
+                )
+                result = ToolPreInvokeResult(
+                    continue_processing=False,
+                    violation=violation,
+                    metadata=metadata,
                 )
-                result = ToolPreInvokeResult(continue_processing=False, violation=violation, metadata=metadata)
 
         else:
             violation = PluginViolation(
-                reason="Tool Check Unavailable", description="Tool arguments check server returned error:", code=f"checkserver_http_status_code:{response.status_code}", details={}
+                reason="Tool Check Unavailable",
+                description="Tool arguments check server returned error:",
+                code=f"checkserver_http_status_code:{response.status_code}",
+                details={},
+            )
+            result = ToolPreInvokeResult(
+                continue_processing=False, violation=violation
             )
-            result = ToolPreInvokeResult(continue_processing=False, violation=violation)
         logger.info(response)
 
         return result
 
-    async def tool_post_invoke(self, payload: ToolPostInvokePayload, context: PluginContext) -> ToolPostInvokeResult:
+    async def tool_post_invoke(
+        self, payload: ToolPostInvokePayload, context: PluginContext
+    ) -> ToolPostInvokeResult:
         """Plugin hook run after a tool is invoked.
 
         Args:
@@ -143,4 +173,84 @@ async def tool_post_invoke(self, payload: ToolPostInvokePayload, context: Plugin
         Returns:
             The result of the plugin's analysis, including whether the tool result should proceed.
         """
-        return ToolPostInvokeResult(continue_processing=True)
+        logger.info(
+            f"[NemoCheck] Starting tool post invoke hook with payload {payload}"
+        )
+
+        # Extract content from payload.result
+        # payload.result format: {'content': [{'type': 'text', 'text': 'Hello, bob!'}]}
+        result_content = payload.result.get("content", [])
+        tool_name = payload.name
+
+        if not result_content:
+            logger.warning(
+                "[NemoCheck] No content in tool result, skipping check"
+            )
+            return ToolPostInvokeResult(continue_processing=True)
+
+        # Extract text content from the content array
+        # TODO: what to do if there's actually multiple texts?
+        text_content = ""
+        for item in result_content:
+            if item.get("type") == "text":
+                text_content += item.get("text", "")
+
+        # Build NeMo check payload for tool response
+        check_nemo_payload = {
+            "model": MODEL_NAME,  # ideally optional
+            "messages": [
+                {"role": "tool", "content": text_content, "name": tool_name}
+            ],
+        }
+
+        logger.debug(
+            f"[NemoCheck] Payload for guardrail check: {check_nemo_payload}"
+        )
+
+        violation = None
+        try:
+            response = requests.post(
+                CHECK_ENDPOINT, headers=headers, json=check_nemo_payload
+            )
+            if response.status_code == 200:
+                data = response.json()
+                status = data.get("status", "blocked")
+                logger.debug(f"[NemoCheck] Rails reply: {data}")
+
+                if status == "success":
+                    metadata = data.get("rails_status")
+                    result = ToolPostInvokeResult(
+                        continue_processing=True, metadata=metadata
+                    )
+                else:  # blocked
+                    metadata = data.get("rails_status")
+                    violation = PluginViolation(
+                        reason=f"Tool response check status: {status}",
+                        description="Rails check blocked tool response",
+                        code=f"checkserver_http_status_code:{response.status_code}",
+                        details=metadata,
+                    )
+                    result = ToolPostInvokeResult(
+                        continue_processing=False,
+                        violation=violation,
+                        metadata=metadata,
+                    )
+            else:
+                violation = PluginViolation(
+                    reason="Tool response check unavailable",
+                    description="Tool response check server returned error",
+                    code=f"checkserver_http_status_code:{response.status_code}",
+                    details={},
+                )
+                result = ToolPostInvokeResult(
+                    continue_processing=False, violation=violation
+                )
+
+            logger.info(f"[NemoCheck] Tool post invoke result: {result}")
+            return result
+
+        except Exception as e:
+            logger.error(f"[NemoCheck] Error checking tool response: {e}")
+            return ToolPostInvokeResult(
+                continue_processing=True
+            )  # Fail open on error
diff --git a/src/server.py b/src/server.py
index 3311b9b..2a1a3cb 100644
--- a/src/server.py
+++ b/src/server.py
@@ -60,7 +60,9 @@ async def getToolPreInvokeResponse(body):
         "tool_args": body["params"]["arguments"],
         "client_session_id": "replaceme",
     }
-    payload = ToolPreInvokePayload(name=body["params"]["name"], args=payload_args)
+    payload = ToolPreInvokePayload(
+        name=body["params"]["name"], args=payload_args
+    )
     # TODO: hard-coded ids
     global_context = GlobalContext(request_id="1", server_id="2")
     logger.debug(f"**** Invoking Tool Pre Invoke with payload: {payload} ****")
@@ -88,7 +90,8 @@ async def getToolPreInvokeResponse(body):
                         ),
                         core.HeaderValueOption(
                             header=core.HeaderValue(
-                                key="x-mcp-denied", raw_value="True".encode("utf-8")
+                                key="x-mcp-denied",
+                                raw_value="True".encode("utf-8"),
                             )
                         ),
                     ],
@@ -103,7 +106,9 @@ async def getToolPreInvokeResponse(body):
             body["params"]["arguments"] = result_payload.args["tool_args"]
             body_mutation = ep.BodyResponse(
                 response=ep.CommonResponse(
-                    body_mutation=ep.BodyMutation(body=json.dumps(body).encode("utf-8"))
+                    body_mutation=ep.BodyMutation(
+                        body=json.dumps(body).encode("utf-8")
+                    )
                 )
             )
         else:
@@ -134,11 +139,32 @@ async def getToolPostInvokeResponse(body):
     )
     logger.info(result)
     if not result.continue_processing:
+        error_body = {
+            "jsonrpc": body["jsonrpc"],
+            "id": body["id"],
+            "error": {"code": -32000, "message": "No go - Tool args forbidden"},
+        }
         body_resp = ep.ProcessingResponse(
             immediate_response=ep.ImmediateResponse(
-                # TODO: hard-coded error reason
-                status=http_status_pb2.HttpStatus(code=http_status_pb2.Forbidden),
-                details="No go",
+                # ok for stream, with error in body
+                status=http_status_pb2.HttpStatus(code=200),
+                headers=ep.HeaderMutation(
+                    set_headers=[
+                        core.HeaderValueOption(
+                            header=core.HeaderValue(
+                                key="content-type",
+                                raw_value="application/json".encode("utf-8"),
+                            )
+                        ),
+                        core.HeaderValueOption(
+                            header=core.HeaderValue(
+                                key="x-mcp-denied",
+                                raw_value="True".encode("utf-8"),
+                            )
+                        ),
+                    ],
+                ),
+                body=(json.dumps(error_body)).encode("utf-8"),
             )
         )
     else:
@@ -147,7 +173,9 @@ async def getToolPostInvokeResponse(body):
             body["result"] = result_payload.result
             body_mutation = ep.BodyResponse(
                 response=ep.CommonResponse(
-                    body_mutation=ep.BodyMutation(body=json.dumps(body).encode("utf-8"))
+                    body_mutation=ep.BodyMutation(
+                        body=json.dumps(body).encode("utf-8")
+                    )
                 )
             )
         else:
@@ -175,7 +203,9 @@ async def getPromptPreFetchResponse(body):
     if not result.continue_processing:
         body_resp = ep.ProcessingResponse(
             immediate_response=ep.ImmediateResponse(
-                status=http_status_pb2.HttpStatus(code=http_status_pb2.Forbidden),
+                status=http_status_pb2.HttpStatus(
+                    code=http_status_pb2.Forbidden
+                ),
                 details="No go",
             )
         )
@@ -184,7 +214,9 @@ async def getPromptPreFetchResponse(body):
         body_resp = ep.ProcessingResponse(
             request_body=ep.BodyResponse(
                 response=ep.CommonResponse(
-                    body_mutation=ep.BodyMutation(body=json.dumps(body).encode("utf-8"))
+                    body_mutation=ep.BodyMutation(
+                        body=json.dumps(body).encode("utf-8")
+                    )
                 )
             )
         )
@@ -287,7 +319,9 @@ async def Process(
                         body = json.loads(text)
                         if "method" in body and body["method"] == "tools/call":
                             body_resp = await getToolPreInvokeResponse(body)
-                        elif "method" in body and body["method"] == "prompts/get":
+                        elif (
+                            "method" in body and body["method"] == "prompts/get"
+                        ):
                             body_resp = await getPromptPreFetchResponse(body)
                         else:
                             body_resp = ep.ProcessingResponse(
@@ -302,7 +336,10 @@ async def Process(
             # ----------------------------------------------------------------
             # Response Body Processing (MCP Tool Results)
             # ----------------------------------------------------------------
-            elif request.HasField("response_body") and request.response_body.body:
+            elif (
+                request.HasField("response_body") and request.response_body.body
+            ):
+                logger.info(f"!!!In Process for response body: {request}")
                 chunk = request.response_body.body
                 resp_body_buf.extend(chunk)
 
@@ -312,15 +349,43 @@ async def Process(
                     except UnicodeDecodeError:
                         logger.debug("Response body not UTF-8; skipping")
                     else:
-                        logger.info(text.split("\n"))
-                        # find data key
-                        data = [d for d in text.split("\n") if d.startswith("data:")]
-                        # logger.info(json.loads(data[0].strip("data:")))
-                        if data:  # List can be empty
-                            data = json.loads(data[0].strip("data:"))
-                        # TODO: check for tool call
-                        if "result" in data and "content" in data["result"]:
-                            body_resp = await getToolPostInvokeResponse(data)
+                        logger.info(f"!!!Text before split: {text.split('\n')}")
+                        # Assume streamable HTTP transport, not SSE - TODO: do we need to check?
+                        lines = [
+                            line.strip()
+                            for line in text.split("\n")
+                            if line.strip()
+                        ]
+
+                        if lines:
+                            try:
+                                # Parse the JSON-RPC response
+                                data = json.loads(lines[0])
+                                logger.info(f"!!!Parsed response: {data}")
+
+                                # Check if this is a tool result response
+                                if (
+                                    "result" in data
+                                    and "content" in data["result"]
+                                ):
+                                    body_resp = await getToolPostInvokeResponse(
+                                        data
+                                    )
+                                else:
+                                    body_resp = ep.ProcessingResponse(
+                                        response_body=ep.BodyResponse(
+                                            response=ep.CommonResponse()
+                                        )
+                                    )
+                            except json.JSONDecodeError as e:
+                                logger.error(
+                                    f"Failed to parse response JSON: {e}"
+                                )
+                                body_resp = ep.ProcessingResponse(
+                                    response_body=ep.BodyResponse(
+                                        response=ep.CommonResponse()
+                                    )
+                                )
                         else:
                             body_resp = ep.ProcessingResponse(
                                 response_body=ep.BodyResponse(

From bc7172fbcb1c6033f89d3ed41aa98b0d5b0434a0 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Tue, 10 Feb 2026 16:44:23 -0700
Subject: [PATCH 02/27] :wrench::art: Update line length and lint

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .pre-commit-config.yaml                       |  3 +-
 plugins/examples/nemo/nemo_wrapper_plugin.py  | 16 +++++++----
 plugins/examples/nemocheck/tests/test_all.py  | 28 ++++++++++++++-----
 .../nemocheck/tests/test_nemocheck.py         |  4 ++-
 4 files changed, 36 insertions(+), 15 deletions(-)

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index a2cd02a..7e76baf 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -4,6 +4,7 @@ repos:
   hooks:
     # Run the linter.
     - id: ruff
-      args: [ --fix ]
+      args: [ --fix, --line-length=80 ]
     # Run the formatter.
     - id: ruff-format
+      args: [ --line-length=80 ]
diff --git a/plugins/examples/nemo/nemo_wrapper_plugin.py b/plugins/examples/nemo/nemo_wrapper_plugin.py
index 047c46f..f7c1a06 100644
--- a/plugins/examples/nemo/nemo_wrapper_plugin.py
+++ b/plugins/examples/nemo/nemo_wrapper_plugin.py
@@ -62,20 +62,24 @@ async def tool_pre_invoke(
                 rails_response = await self._rails.generate_async(
                     messages=[{"role": "user", "content": payload_args}]
                 )
-            except (
-                asyncio.CancelledError
-            ):  # asyncio.exceptions.CancelledError is thrown by nemo, need to catch
-                logging.exception("An error occurred in the nemo plugin except block:")
+            except asyncio.CancelledError:  # asyncio.exceptions.CancelledError is thrown by nemo, need to catch
+                logging.exception(
+                    "An error occurred in the nemo plugin except block:"
+                )
             finally:
                 logger.warning("[NemoWrapperPlugin] Async rails executed")
                 logger.warning(rails_response)
         if rails_response and "PII detected" in rails_response["content"]:
-            logger.warning("[NemoWrapperPlugin] PII detected, stopping processing")
+            logger.warning(
+                "[NemoWrapperPlugin] PII detected, stopping processing"
+            )
             return ToolPreInvokeResult(
                 modified_payload=payload, continue_processing=False
             )
         logger.warning("[NemoWrapperPlugin] No PII detected, continuing")
-        return ToolPreInvokeResult(modified_payload=payload, continue_processing=True)
+        return ToolPreInvokeResult(
+            modified_payload=payload, continue_processing=True
+        )
 
     async def tool_post_invoke(
         self, payload: ToolPostInvokePayload, context: PluginContext
diff --git a/plugins/examples/nemocheck/tests/test_all.py b/plugins/examples/nemocheck/tests/test_all.py
index 85d1c22..f2263e5 100644
--- a/plugins/examples/nemocheck/tests/test_all.py
+++ b/plugins/examples/nemocheck/tests/test_all.py
@@ -33,9 +33,13 @@ def plugin_manager():
 async def test_prompt_pre_hook(plugin_manager: PluginManager):
     """Test prompt pre hook across all registered plugins."""
     # Customize payload for testing
-    payload = PromptPrehookPayload(prompt_id="test_prompt", args={"arg0": "This is an argument"})
+    payload = PromptPrehookPayload(
+        prompt_id="test_prompt", args={"arg0": "This is an argument"}
+    )
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(PromptHookType.PROMPT_PRE_FETCH, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        PromptHookType.PROMPT_PRE_FETCH, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing
 
@@ -44,11 +48,17 @@ async def test_prompt_pre_hook(plugin_manager: PluginManager):
 async def test_prompt_post_hook(plugin_manager: PluginManager):
     """Test prompt post hook across all registered plugins."""
     # Customize payload for testing
-    message = Message(content=TextContent(type="text", text="prompt"), role=Role.USER)
+    message = Message(
+        content=TextContent(type="text", text="prompt"), role=Role.USER
+    )
     prompt_result = PromptResult(messages=[message])
-    payload = PromptPosthookPayload(prompt_id="test_prompt", result=prompt_result)
+    payload = PromptPosthookPayload(
+        prompt_id="test_prompt", result=prompt_result
+    )
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(PromptHookType.PROMPT_POST_FETCH, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        PromptHookType.PROMPT_POST_FETCH, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing
 
@@ -57,8 +67,12 @@ async def test_prompt_post_hook(plugin_manager: PluginManager):
 async def test_tool_post_hook(plugin_manager: PluginManager):
     """Test tool post hook across all registered plugins."""
     # Customize payload for testing
-    payload = ToolPostInvokePayload(name="test_tool", result={"output0": "output value"})
+    payload = ToolPostInvokePayload(
+        name="test_tool", result={"output0": "output value"}
+    )
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(ToolHookType.TOOL_POST_INVOKE, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        ToolHookType.TOOL_POST_INVOKE, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 4321c51..6ee310e 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -25,7 +25,9 @@ async def test_nemocheck():
     plugin = NemoCheck(config)
 
     # Test your plugin logic
-    payload = PromptPrehookPayload(prompt_id="test_prompt", args={"arg0": "This is an argument"})
+    payload = PromptPrehookPayload(
+        prompt_id="test_prompt", args={"arg0": "This is an argument"}
+    )
     context = GlobalContext(request_id="1")
     result = await plugin.prompt_pre_fetch(payload, context)
     assert result.continue_processing

From 51654236dfa36339c4b5a7fc084c5459be7a7422 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Tue, 10 Feb 2026 16:45:54 -0700
Subject: [PATCH 03/27] :rewind: Revert port change

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/k8deploy/deploy.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/plugins/examples/nemocheck/k8deploy/deploy.yaml b/plugins/examples/nemocheck/k8deploy/deploy.yaml
index 0971cb8..4ae0703 100644
--- a/plugins/examples/nemocheck/k8deploy/deploy.yaml
+++ b/plugins/examples/nemocheck/k8deploy/deploy.yaml
@@ -34,7 +34,7 @@ spec:
             - name: MODEL_NAME # Currently only for logging.
               value: "meta-llama/llama-3-3-70b-instruct"
             - name: CHECK_ENDPOINT
-              value: "http://nemo-guardrails-service:50053/v1/guardrail/checks"
+              value: "http://nemo-guardrails-service:8000/v1/guardrail/checks"
             - name: PLUGINS_SERVER_HOST
               value: "0.0.0.0"
             - name: PLUGINS_ENABLED

From 188e8ecc02a0775a8077eaf6f0794bef5d9acc96 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Tue, 10 Feb 2026 16:53:44 -0700
Subject: [PATCH 04/27] :sparkles: SSE format in tools

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 71 +++++++++++++++++++++++++++++----------------------
 1 file changed, 40 insertions(+), 31 deletions(-)

diff --git a/src/server.py b/src/server.py
index 2a1a3cb..af28cc5 100644
--- a/src/server.py
+++ b/src/server.py
@@ -350,37 +350,46 @@ async def Process(
                         logger.debug("Response body not UTF-8; skipping")
                     else:
                         logger.info(f"!!!Text before split: {text.split('\n')}")
-                        # Assume streamable HTTP transport, not SSE - TODO: do we need to check?
-                        lines = [
-                            line.strip()
-                            for line in text.split("\n")
-                            if line.strip()
-                        ]
-
-                        if lines:
-                            try:
-                                # Parse the JSON-RPC response
-                                data = json.loads(lines[0])
-                                logger.info(f"!!!Parsed response: {data}")
-
-                                # Check if this is a tool result response
-                                if (
-                                    "result" in data
-                                    and "content" in data["result"]
-                                ):
-                                    body_resp = await getToolPostInvokeResponse(
-                                        data
-                                    )
-                                else:
-                                    body_resp = ep.ProcessingResponse(
-                                        response_body=ep.BodyResponse(
-                                            response=ep.CommonResponse()
-                                        )
-                                    )
-                            except json.JSONDecodeError as e:
-                                logger.error(
-                                    f"Failed to parse response JSON: {e}"
-                                )
+
+                        # Handle both SSE format and plain JSON-RPC format
+                        data = None
+
+                        # Check if this is SSE format (starts with "event:" or "data:")
+                        if text.strip().startswith(("event:", "data:")):
+                            # Parse SSE format
+                            lines = text.split("\n")
+                            for line in lines:
+                                line = line.strip()
+                                if line.startswith("data:"):
+                                    json_str = line[5:].strip()  # Remove "data:" prefix
+                                    try:
+                                        data = json.loads(json_str)
+                                        break
+                                    except json.JSONDecodeError:
+                                        continue
+                        else:
+                            # Parse plain JSON-RPC format
+                            lines = [
+                                line.strip()
+                                for line in text.split("\n")
+                                if line.strip()
+                            ]
+                            if lines:
+                                try:
+                                    data = json.loads(lines[0])
+                                except json.JSONDecodeError as e:
+                                    logger.error(f"Failed to parse JSON: {e}")
+
+                        if data:
+                            logger.info(f"!!!Parsed response: {data}")
+
+                            # Check if this is a tool result response
+                            if (
+                                "result" in data
+                                and "content" in data["result"]
+                            ):
+                                body_resp = await getToolPostInvokeResponse(data)
+                            else:
                                 body_resp = ep.ProcessingResponse(
                                     response_body=ep.BodyResponse(
                                         response=ep.CommonResponse()

From 7cfad9eb33b675a88705a1aa0dc49a2a6d45305f Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Wed, 11 Feb 2026 07:42:00 -0700
Subject: [PATCH 05/27] :loud_sound: Update tool response error message and
 debug logs

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/server.py b/src/server.py
index af28cc5..afe6368 100644
--- a/src/server.py
+++ b/src/server.py
@@ -142,7 +142,7 @@ async def getToolPostInvokeResponse(body):
         error_body = {
             "jsonrpc": body["jsonrpc"],
             "id": body["id"],
-            "error": {"code": -32000, "message": "No go - Tool args forbidden"},
+            "error": {"code": -32000, "message": "Tool response forbidden"},
         }
         body_resp = ep.ProcessingResponse(
             immediate_response=ep.ImmediateResponse(
@@ -339,7 +339,7 @@ async def Process(
             elif (
                 request.HasField("response_body") and request.response_body.body
             ):
-                logger.info(f"!!!In Process for response body: {request}")
+                logger.debug(f"Processing response body: {request}")
                 chunk = request.response_body.body
                 resp_body_buf.extend(chunk)
 
@@ -349,7 +349,7 @@ async def Process(
                     except UnicodeDecodeError:
                         logger.debug("Response body not UTF-8; skipping")
                     else:
-                        logger.info(f"!!!Text before split: {text.split('\n')}")
+                        logger.debug(f"Response body text: {text.split('\n')}")
 
                         # Handle both SSE format and plain JSON-RPC format
                         data = None
@@ -381,7 +381,7 @@ async def Process(
                                     logger.error(f"Failed to parse JSON: {e}")
 
                         if data:
-                            logger.info(f"!!!Parsed response: {data}")
+                            logger.debug(f"Parsed response data: {data}")
 
                             # Check if this is a tool result response
                             if (

From 67dccea5838ce94a9254ff7bf819b7bcb09253f8 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Wed, 11 Feb 2026 09:47:28 -0700
Subject: [PATCH 06/27] :goal_net: Buffer intermediate response chunks

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/src/server.py b/src/server.py
index afe6368..b175825 100644
--- a/src/server.py
+++ b/src/server.py
@@ -344,6 +344,7 @@ async def Process(
                 resp_body_buf.extend(chunk)
 
                 if getattr(request.response_body, "end_of_stream", False):
+                    # End of stream reached - process complete buffered response
                     try:
                         text = resp_body_buf.decode("utf-8")
                     except UnicodeDecodeError:
@@ -403,6 +404,16 @@ async def Process(
                             )
                         yield body_resp
                     resp_body_buf.clear()
+                else:
+                    # Intermediate chunk - buffer only, don't process yet
+                    # TODO: how should this be handled?
+                    logger.debug(f"Buffering intermediate chunk ({len(chunk)} bytes), waiting for end_of_stream")
+                    # Yield empty response to acknowledge chunk receipt
+                    yield ep.ProcessingResponse(
+                        response_body=ep.BodyResponse(
+                            response=ep.CommonResponse()
+                        )
+                    )
             # ----------------------------------------------------------------
             # Response Body Processing (No body field)
             # ----------------------------------------------------------------

From 0b3e4b16b8ea8a21e0b5cc83307cfaeb6e469424 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Wed, 11 Feb 2026 10:39:16 -0700
Subject: [PATCH 07/27] :goal_net: Handle empty chunk cases with EOS

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 198 +++++++++++++++++++++++++++-----------------------
 1 file changed, 106 insertions(+), 92 deletions(-)

diff --git a/src/server.py b/src/server.py
index b175825..88fc7a7 100644
--- a/src/server.py
+++ b/src/server.py
@@ -137,34 +137,37 @@ async def getToolPostInvokeResponse(body):
     result, _ = await manager.invoke_hook(
         ToolHookType.TOOL_POST_INVOKE, payload, global_context=global_context
     )
-    logger.info(result)
+    logger.debug(f"**** Tool Post Invoke result {result}")
     if not result.continue_processing:
+        # Build error response to replace the tool result
         error_body = {
             "jsonrpc": body["jsonrpc"],
             "id": body["id"],
             "error": {"code": -32000, "message": "Tool response forbidden"},
         }
         body_resp = ep.ProcessingResponse(
-            immediate_response=ep.ImmediateResponse(
-                # ok for stream, with error in body
-                status=http_status_pb2.HttpStatus(code=200),
-                headers=ep.HeaderMutation(
-                    set_headers=[
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="content-type",
-                                raw_value="application/json".encode("utf-8"),
-                            )
-                        ),
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="x-mcp-denied",
-                                raw_value="True".encode("utf-8"),
-                            )
-                        ),
-                    ],
-                ),
-                body=(json.dumps(error_body)).encode("utf-8"),
+            response_body=ep.BodyResponse(
+                response=ep.CommonResponse(
+                    header_mutation=ep.HeaderMutation(
+                        set_headers=[
+                            core.HeaderValueOption(
+                                header=core.HeaderValue(
+                                    key="content-type",
+                                    raw_value="application/json".encode("utf-8"),
+                                )
+                            ),
+                            core.HeaderValueOption(
+                                header=core.HeaderValue(
+                                    key="x-mcp-denied",
+                                    raw_value="True".encode("utf-8"),
+                                )
+                            ),
+                        ],
+                    ),
+                    body_mutation=ep.BodyMutation(
+                        body=(json.dumps(error_body)).encode("utf-8")
+                    ),
+                )
             )
         )
     else:
@@ -180,7 +183,7 @@ async def getToolPostInvokeResponse(body):
             )
         else:
             body_mutation = ep.BodyResponse(response=ep.CommonResponse())
-        body_resp = ep.ProcessingResponse(request_body=body_mutation)
+        body_resp = ep.ProcessingResponse(response_body=body_mutation)
     return body_resp
 
 
@@ -336,94 +339,105 @@ async def Process(
             # ----------------------------------------------------------------
             # Response Body Processing (MCP Tool Results)
             # ----------------------------------------------------------------
-            elif (
-                request.HasField("response_body") and request.response_body.body
-            ):
+            elif request.HasField("response_body"):
                 logger.debug(f"Processing response body: {request}")
-                chunk = request.response_body.body
-                resp_body_buf.extend(chunk)
 
+                # Buffer content if present in this chunk
+                if request.response_body.body:
+                    chunk = request.response_body.body
+                    resp_body_buf.extend(chunk)
+                    logger.debug(f"Buffered chunk ({len(chunk)} bytes)")
+
+                # Check for end of stream (regardless of whether this chunk has content)
                 if getattr(request.response_body, "end_of_stream", False):
-                    # End of stream reached - process complete buffered response
-                    try:
-                        text = resp_body_buf.decode("utf-8")
-                    except UnicodeDecodeError:
-                        logger.debug("Response body not UTF-8; skipping")
-                    else:
-                        logger.debug(f"Response body text: {text.split('\n')}")
-
-                        # Handle both SSE format and plain JSON-RPC format
-                        data = None
-
-                        # Check if this is SSE format (starts with "event:" or "data:")
-                        if text.strip().startswith(("event:", "data:")):
-                            # Parse SSE format
-                            lines = text.split("\n")
-                            for line in lines:
-                                line = line.strip()
-                                if line.startswith("data:"):
-                                    json_str = line[5:].strip()  # Remove "data:" prefix
-                                    try:
-                                        data = json.loads(json_str)
-                                        break
-                                    except json.JSONDecodeError:
-                                        continue
+                    logger.debug("End of stream reached, processing complete buffered response")
+
+                    # Process buffered content if any
+                    if resp_body_buf:
+                        try:
+                            text = resp_body_buf.decode("utf-8")
+                        except UnicodeDecodeError:
+                            logger.debug("Response body not UTF-8; skipping")
+                            body_resp = ep.ProcessingResponse(
+                                response_body=ep.BodyResponse(
+                                    response=ep.CommonResponse()
+                                )
+                            )
                         else:
-                            # Parse plain JSON-RPC format
-                            lines = [
-                                line.strip()
-                                for line in text.split("\n")
-                                if line.strip()
-                            ]
-                            if lines:
-                                try:
-                                    data = json.loads(lines[0])
-                                except json.JSONDecodeError as e:
-                                    logger.error(f"Failed to parse JSON: {e}")
-
-                        if data:
-                            logger.debug(f"Parsed response data: {data}")
-
-                            # Check if this is a tool result response
-                            if (
-                                "result" in data
-                                and "content" in data["result"]
-                            ):
-                                body_resp = await getToolPostInvokeResponse(data)
+                            logger.debug(f"Response body text: {text.split('\n')}")
+
+                            # Handle both SSE format and plain JSON-RPC format
+                            data = None
+
+                            # Check if this is SSE format (starts with "event:" or "data:")
+                            if text.strip().startswith(("event:", "data:")):
+                                # Parse SSE format
+                                lines = text.split("\n")
+                                for line in lines:
+                                    line = line.strip()
+                                    if line.startswith("data:"):
+                                        json_str = line[5:].strip()  # Remove "data:" prefix
+                                        logger.debug(f"Extracted JSON from SSE: {json_str}")
+                                        try:
+                                            data = json.loads(json_str)
+                                            break
+                                        except json.JSONDecodeError:
+                                            continue
+                            else:
+                                # Parse plain JSON-RPC format
+                                lines = [
+                                    line.strip()
+                                    for line in text.split("\n")
+                                    if line.strip()
+                                ]
+                                if lines:
+                                    try:
+                                        data = json.loads(lines[0])
+                                    except json.JSONDecodeError as e:
+                                        logger.error(f"Failed to parse JSON: {e}")
+
+                            if data:
+                                logger.debug(f"Parsed response data: {data}")
+
+                                # Check if this is a tool result response
+                                if (
+                                    "result" in data
+                                    and "content" in data["result"]
+                                ):
+                                    logger.info("Invoking tool post-invoke hook")
+                                    body_resp = await getToolPostInvokeResponse(data)
+                                else:
+                                    body_resp = ep.ProcessingResponse(
+                                        response_body=ep.BodyResponse(
+                                            response=ep.CommonResponse()
+                                        )
+                                    )
                             else:
+                                logger.warning("No data parsed from response body")
                                 body_resp = ep.ProcessingResponse(
                                     response_body=ep.BodyResponse(
                                         response=ep.CommonResponse()
                                     )
                                 )
-                        else:
-                            body_resp = ep.ProcessingResponse(
-                                response_body=ep.BodyResponse(
-                                    response=ep.CommonResponse()
-                                )
+                    else:
+                        # Empty buffer at end of stream
+                        logger.debug("End of stream with empty buffer")
+                        body_resp = ep.ProcessingResponse(
+                            response_body=ep.BodyResponse(
+                                response=ep.CommonResponse()
                             )
-                        yield body_resp
+                        )
+
+                    yield body_resp
                     resp_body_buf.clear()
                 else:
-                    # Intermediate chunk - buffer only, don't process yet
-                    # TODO: how should this be handled?
-                    logger.debug(f"Buffering intermediate chunk ({len(chunk)} bytes), waiting for end_of_stream")
-                    # Yield empty response to acknowledge chunk receipt
+                    # Intermediate chunk - acknowledge but don't process yet
+                    logger.debug(f"Buffering intermediate chunk, waiting for end_of_stream")
                     yield ep.ProcessingResponse(
                         response_body=ep.BodyResponse(
                             response=ep.CommonResponse()
                         )
                     )
-            # ----------------------------------------------------------------
-            # Response Body Processing (No body field)
-            # ----------------------------------------------------------------
-            elif request.HasField("response_body"):
-                logger.warning("On Response, no body.")
-                logger.warning(request)
-                body_resp = ep.ProcessingResponse(
-                    response_body=ep.BodyResponse(response=ep.CommonResponse())
-                )
-                yield body_resp
 
             else:
                 # Unhandled request types

From e33d36e0151de8b4a32b06a6592e6d22b52178ca Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Wed, 11 Feb 2026 10:58:35 -0700
Subject: [PATCH 08/27] :white_check_mark: Add nemocheck plugin tool pre and
 tool post invoke test cases

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .github/workflows/ci.yaml                     |   7 +-
 .../nemocheck/tests/test_nemocheck.py         | 173 +++++++++++++++++-
 2 files changed, 169 insertions(+), 11 deletions(-)

diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 7536ae2..9c50347 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -47,8 +47,9 @@ jobs:
         run: |
           echo "Running nemocheck ext plugin tests.."
           uv sync --all-groups
+
+      # Tests
       - name: Run nemocheck external plugin tests
         working-directory: ./plugins/examples/nemocheck
-        run: uv run pytest tests 
-          
-      # - name: Test
+        run: uv run pytest tests
+
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 6ee310e..10d7341 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -1,5 +1,8 @@
 """Tests for plugin."""
 
+# Standard
+from unittest.mock import Mock, patch
+
 # Third-Party
 import pytest
 
@@ -9,25 +12,179 @@
     PluginConfig,
     GlobalContext,
     PromptPrehookPayload,
+    ToolPostInvokePayload,
+    ToolPreInvokePayload,
 )
 
 
-@pytest.mark.asyncio
-async def test_nemocheck():
-    """Test plugin prompt prefetch hook."""
+@pytest.fixture
+def plugin():
+    """Create a NemoCheck plugin instance."""
     config = PluginConfig(
         name="test",
         kind="nemocheck.NemoCheck",
-        hooks=["prompt_pre_fetch"],
-        config={"setting_one": "test_value"},
+        hooks=["prompt_pre_fetch", "tool_pre_invoke", "tool_post_invoke"],
+        config={},
     )
+    return NemoCheck(config)
+
+
+@pytest.fixture
+def context():
+    """Create a GlobalContext instance."""
+    return GlobalContext(request_id="1")
+
+
+def mock_http_response(status_code, response_data=None):
+    """Helper to create mock HTTP responses."""
+    mock_response = Mock()
+    mock_response.status_code = status_code
+    if response_data:
+        mock_response.json.return_value = response_data
+    return mock_response
 
-    plugin = NemoCheck(config)
 
-    # Test your plugin logic
+@pytest.mark.asyncio
+async def test_prompt_pre_fetch(plugin, context):
+    """Test plugin prompt prefetch hook."""
     payload = PromptPrehookPayload(
         prompt_id="test_prompt", args={"arg0": "This is an argument"}
     )
-    context = GlobalContext(request_id="1")
     result = await plugin.prompt_pre_fetch(payload, context)
     assert result.continue_processing
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "status_code,response_data,expected_continue,has_violation",
+    [
+        (200, {"status": "success", "rails_status": {"detect senstitive data": {"status": "success"}}}, True, False),
+        (200, {"status": "blocked", "rails_status": {"detect hap": {"status": "blocked"}}}, False, True),
+        (503, None, False, True),
+    ],
+)
+async def test_tool_pre_invoke_scenarios(
+    plugin, context, status_code, response_data, expected_continue, has_violation
+):
+    """Test tool_pre_invoke with various scenarios."""
+    payload = ToolPreInvokePayload(
+        name="test_tool",
+        args={"tool_args": '{"param": "value"}'},
+    )
+
+    with patch(
+        "nemocheck.plugin.requests.post",
+        return_value=mock_http_response(status_code, response_data),
+    ):
+        result = await plugin.tool_pre_invoke(payload, context)
+
+    assert result.continue_processing == expected_continue
+    assert (result.violation is not None) == has_violation
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "status_code,response_data,expected_continue,has_violation",
+    [
+        (200, {"status": "success", "rails_status": {"detect senstitive data": {"status": "success"}}}, True, False),
+        (200, {"status": "blocked", "rails_status": {"detect hap": {"status": "blocked"}}}, False, True),
+        (500, None, False, True),
+    ],
+)
+async def test_tool_post_invoke_http_scenarios(
+    plugin, context, status_code, response_data, expected_continue, has_violation
+):
+    """Test tool_post_invoke with various HTTP response scenarios."""
+    payload = ToolPostInvokePayload(
+        name="test_tool",
+        result={"content": [{"type": "text", "text": "Test content"}]},
+    )
+
+    with patch(
+        "nemocheck.plugin.requests.post",
+        return_value=mock_http_response(status_code, response_data),
+    ):
+        result = await plugin.tool_post_invoke(payload, context)
+
+    assert result.continue_processing == expected_continue
+    assert (result.violation is not None) == has_violation
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "result_data,should_continue",
+    [
+        ({"content": []}, True),  # Empty content
+        ({"output": "value"}, True),  # No content key
+    ],
+)
+async def test_tool_post_invoke_passthrough_content_cases(plugin, context, result_data, should_continue):
+    """Test tool_post_invoke no/empty content cases that do not flag."""
+    payload = ToolPostInvokePayload(name="test_tool", result=result_data)
+    result = await plugin.tool_post_invoke(payload, context)
+    assert result.continue_processing == should_continue
+    assert result.violation is None
+
+
+@pytest.mark.asyncio
+async def test_tool_post_invoke_concatenates_text(plugin, context):
+    """Test tool_post_invoke concatenates multiple text items."""
+    payload = ToolPostInvokePayload(
+        name="test_tool",
+        result={
+            "content": [
+                {"type": "text", "text": "First. "},
+                {"type": "text", "text": "Second."},
+            ]
+        },
+    )
+
+    with patch(
+        "nemocheck.plugin.requests.post",
+        return_value=mock_http_response(200, {"status": "success", "rails_status": {}}),
+    ) as mock_post:
+        result = await plugin.tool_post_invoke(payload, context)
+
+    assert result.continue_processing
+    sent_content = mock_post.call_args[1]["json"]["messages"][0]["content"]
+    assert sent_content == "First. Second."
+
+
+@pytest.mark.asyncio
+async def test_tool_post_invoke_filters_non_text(plugin, context):
+    """Test tool_post_invoke filters non-text content."""
+    payload = ToolPostInvokePayload(
+        name="test_tool",
+        result={
+            "content": [
+                {"type": "image", "url": "http://example.com/img.png"},
+                {"type": "text", "text": "Text only"},
+            ]
+        },
+    )
+
+    with patch(
+        "nemocheck.plugin.requests.post",
+        return_value=mock_http_response(200, {"status": "success", "rails_status": {}}),
+    ) as mock_post:
+        result = await plugin.tool_post_invoke(payload, context)
+
+    assert result.continue_processing
+    sent_content = mock_post.call_args[1]["json"]["messages"][0]["content"]
+    assert sent_content == "Text only"
+
+
+@pytest.mark.asyncio
+async def test_tool_post_invoke_fails_open_on_exception(plugin, context):
+    """Test tool_post_invoke fails open on exceptions."""
+    payload = ToolPostInvokePayload(
+        name="test_tool",
+        result={"content": [{"type": "text", "text": "content"}]},
+    )
+
+    with patch("nemocheck.plugin.requests.post", side_effect=Exception("Network error")):
+        result = await plugin.tool_post_invoke(payload, context)
+
+    assert result.continue_processing
+    assert result.violation is None
+

From f8807150425c8b088d97a7bf86e7c59e7002a23e Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Wed, 11 Feb 2026 11:09:24 -0700
Subject: [PATCH 09/27] :white_check_mark: Add initial server tests

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .github/workflows/ci.yaml                     |  13 +-
 .../nemocheck/tests/test_nemocheck.py         |  75 +++++-
 src/__init__.py                               |   0
 src/server.py                                 |  39 ++-
 tests/__init__.py                             |   0
 tests/pytest.ini                              |  11 +
 tests/test_server.py                          | 232 ++++++++++++++++++
 7 files changed, 347 insertions(+), 23 deletions(-)
 create mode 100644 src/__init__.py
 create mode 100644 tests/__init__.py
 create mode 100644 tests/pytest.ini
 create mode 100644 tests/test_server.py

diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 9c50347..b3937ba 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -42,14 +42,23 @@ jobs:
             git diff --stat
             exit 1
           }
+
+      # Plugin tests
       - name: Install nemocheck external plugin dependencies
         working-directory: ./plugins/examples/nemocheck
         run: |
           echo "Running nemocheck ext plugin tests.."
           uv sync --all-groups
-
-      # Tests
       - name: Run nemocheck external plugin tests
         working-directory: ./plugins/examples/nemocheck
         run: uv run pytest tests
 
+      # Server tests
+      - name: Install server test dependencies
+        run: |
+          echo "Installing pytest and dependencies for server tests..."
+          pip install pytest pytest-asyncio
+      - name: Run server unit tests
+        run: |
+          echo "Running server unit tests..."
+          uv run pytest tests
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 10d7341..a0d59b6 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -58,13 +58,36 @@ async def test_prompt_pre_fetch(plugin, context):
 @pytest.mark.parametrize(
     "status_code,response_data,expected_continue,has_violation",
     [
-        (200, {"status": "success", "rails_status": {"detect senstitive data": {"status": "success"}}}, True, False),
-        (200, {"status": "blocked", "rails_status": {"detect hap": {"status": "blocked"}}}, False, True),
+        (
+            200,
+            {
+                "status": "success",
+                "rails_status": {
+                    "detect senstitive data": {"status": "success"}
+                },
+            },
+            True,
+            False,
+        ),
+        (
+            200,
+            {
+                "status": "blocked",
+                "rails_status": {"detect hap": {"status": "blocked"}},
+            },
+            False,
+            True,
+        ),
         (503, None, False, True),
     ],
 )
 async def test_tool_pre_invoke_scenarios(
-    plugin, context, status_code, response_data, expected_continue, has_violation
+    plugin,
+    context,
+    status_code,
+    response_data,
+    expected_continue,
+    has_violation,
 ):
     """Test tool_pre_invoke with various scenarios."""
     payload = ToolPreInvokePayload(
@@ -86,13 +109,36 @@ async def test_tool_pre_invoke_scenarios(
 @pytest.mark.parametrize(
     "status_code,response_data,expected_continue,has_violation",
     [
-        (200, {"status": "success", "rails_status": {"detect senstitive data": {"status": "success"}}}, True, False),
-        (200, {"status": "blocked", "rails_status": {"detect hap": {"status": "blocked"}}}, False, True),
+        (
+            200,
+            {
+                "status": "success",
+                "rails_status": {
+                    "detect senstitive data": {"status": "success"}
+                },
+            },
+            True,
+            False,
+        ),
+        (
+            200,
+            {
+                "status": "blocked",
+                "rails_status": {"detect hap": {"status": "blocked"}},
+            },
+            False,
+            True,
+        ),
         (500, None, False, True),
     ],
 )
 async def test_tool_post_invoke_http_scenarios(
-    plugin, context, status_code, response_data, expected_continue, has_violation
+    plugin,
+    context,
+    status_code,
+    response_data,
+    expected_continue,
+    has_violation,
 ):
     """Test tool_post_invoke with various HTTP response scenarios."""
     payload = ToolPostInvokePayload(
@@ -118,7 +164,9 @@ async def test_tool_post_invoke_http_scenarios(
         ({"output": "value"}, True),  # No content key
     ],
 )
-async def test_tool_post_invoke_passthrough_content_cases(plugin, context, result_data, should_continue):
+async def test_tool_post_invoke_passthrough_content_cases(
+    plugin, context, result_data, should_continue
+):
     """Test tool_post_invoke no/empty content cases that do not flag."""
     payload = ToolPostInvokePayload(name="test_tool", result=result_data)
     result = await plugin.tool_post_invoke(payload, context)
@@ -141,7 +189,9 @@ async def test_tool_post_invoke_concatenates_text(plugin, context):
 
     with patch(
         "nemocheck.plugin.requests.post",
-        return_value=mock_http_response(200, {"status": "success", "rails_status": {}}),
+        return_value=mock_http_response(
+            200, {"status": "success", "rails_status": {}}
+        ),
     ) as mock_post:
         result = await plugin.tool_post_invoke(payload, context)
 
@@ -165,7 +215,9 @@ async def test_tool_post_invoke_filters_non_text(plugin, context):
 
     with patch(
         "nemocheck.plugin.requests.post",
-        return_value=mock_http_response(200, {"status": "success", "rails_status": {}}),
+        return_value=mock_http_response(
+            200, {"status": "success", "rails_status": {}}
+        ),
     ) as mock_post:
         result = await plugin.tool_post_invoke(payload, context)
 
@@ -182,9 +234,10 @@ async def test_tool_post_invoke_fails_open_on_exception(plugin, context):
         result={"content": [{"type": "text", "text": "content"}]},
     )
 
-    with patch("nemocheck.plugin.requests.post", side_effect=Exception("Network error")):
+    with patch(
+        "nemocheck.plugin.requests.post", side_effect=Exception("Network error")
+    ):
         result = await plugin.tool_post_invoke(payload, context)
 
     assert result.continue_processing
     assert result.violation is None
-
diff --git a/src/__init__.py b/src/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/server.py b/src/server.py
index 88fc7a7..4dac2aa 100644
--- a/src/server.py
+++ b/src/server.py
@@ -153,7 +153,9 @@ async def getToolPostInvokeResponse(body):
                             core.HeaderValueOption(
                                 header=core.HeaderValue(
                                     key="content-type",
-                                    raw_value="application/json".encode("utf-8"),
+                                    raw_value="application/json".encode(
+                                        "utf-8"
+                                    ),
                                 )
                             ),
                             core.HeaderValueOption(
@@ -350,7 +352,9 @@ async def Process(
 
                 # Check for end of stream (regardless of whether this chunk has content)
                 if getattr(request.response_body, "end_of_stream", False):
-                    logger.debug("End of stream reached, processing complete buffered response")
+                    logger.debug(
+                        "End of stream reached, processing complete buffered response"
+                    )
 
                     # Process buffered content if any
                     if resp_body_buf:
@@ -364,7 +368,8 @@ async def Process(
                                 )
                             )
                         else:
-                            logger.debug(f"Response body text: {text.split('\n')}")
+                            lines = text.split("\n")
+                            logger.debug(f"Response body text: {lines}")
 
                             # Handle both SSE format and plain JSON-RPC format
                             data = None
@@ -376,8 +381,12 @@ async def Process(
                                 for line in lines:
                                     line = line.strip()
                                     if line.startswith("data:"):
-                                        json_str = line[5:].strip()  # Remove "data:" prefix
-                                        logger.debug(f"Extracted JSON from SSE: {json_str}")
+                                        json_str = line[
+                                            5:
+                                        ].strip()  # Remove "data:" prefix
+                                        logger.debug(
+                                            f"Extracted JSON from SSE: {json_str}"
+                                        )
                                         try:
                                             data = json.loads(json_str)
                                             break
@@ -394,7 +403,9 @@ async def Process(
                                     try:
                                         data = json.loads(lines[0])
                                     except json.JSONDecodeError as e:
-                                        logger.error(f"Failed to parse JSON: {e}")
+                                        logger.error(
+                                            f"Failed to parse JSON: {e}"
+                                        )
 
                             if data:
                                 logger.debug(f"Parsed response data: {data}")
@@ -404,8 +415,12 @@ async def Process(
                                     "result" in data
                                     and "content" in data["result"]
                                 ):
-                                    logger.info("Invoking tool post-invoke hook")
-                                    body_resp = await getToolPostInvokeResponse(data)
+                                    logger.info(
+                                        "Invoking tool post-invoke hook"
+                                    )
+                                    body_resp = await getToolPostInvokeResponse(
+                                        data
+                                    )
                                 else:
                                     body_resp = ep.ProcessingResponse(
                                         response_body=ep.BodyResponse(
@@ -413,7 +428,9 @@ async def Process(
                                         )
                                     )
                             else:
-                                logger.warning("No data parsed from response body")
+                                logger.warning(
+                                    "No data parsed from response body"
+                                )
                                 body_resp = ep.ProcessingResponse(
                                     response_body=ep.BodyResponse(
                                         response=ep.CommonResponse()
@@ -432,7 +449,9 @@ async def Process(
                     resp_body_buf.clear()
                 else:
                     # Intermediate chunk - acknowledge but don't process yet
-                    logger.debug(f"Buffering intermediate chunk, waiting for end_of_stream")
+                    logger.debug(
+                        "Buffering intermediate chunk, waiting for end_of_stream"
+                    )
                     yield ep.ProcessingResponse(
                         response_body=ep.BodyResponse(
                             response=ep.CommonResponse()
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/pytest.ini b/tests/pytest.ini
new file mode 100644
index 0000000..9a43301
--- /dev/null
+++ b/tests/pytest.ini
@@ -0,0 +1,11 @@
+[pytest]
+log_cli = false
+log_cli_level = INFO
+log_cli_format = %(asctime)s [%(module)s] [%(levelname)s] %(message)s
+log_cli_date_format = %Y-%m-%d %H:%M:%S
+log_level = INFO
+log_format = %(asctime)s [%(module)s] [%(levelname)s] %(message)s
+log_date_format = %Y-%m-%d %H:%M:%S
+pythonpath = . src
+filterwarnings =
+    ignore::DeprecationWarning
\ No newline at end of file
diff --git a/tests/test_server.py b/tests/test_server.py
new file mode 100644
index 0000000..0a5f033
--- /dev/null
+++ b/tests/test_server.py
@@ -0,0 +1,232 @@
+"""Unit tests for ext-proc server functions
+
+These tests use dynamic import and mocking to avoid proto dependencies.
+"""
+
+# Standard
+from unittest.mock import AsyncMock, Mock, MagicMock
+import sys
+
+# Third-Party
+import pytest
+
+# First-Party
+from mcpgateway.plugins.framework import (
+    ToolPostInvokeResult,
+    ToolPostInvokePayload,
+    PluginViolation,
+)
+
+
+@pytest.fixture
+def mock_envoy_modules():
+    """Mock envoy protobuf modules to avoid proto dependencies."""
+    # Create mock modules
+    mock_ep = MagicMock()
+    mock_ep_grpc = MagicMock()
+    mock_core = MagicMock()
+    mock_http_status = MagicMock()
+
+    # Add to sys.modules before importing server
+    sys.modules["envoy"] = MagicMock()
+    sys.modules["envoy.service"] = MagicMock()
+    sys.modules["envoy.service.ext_proc"] = MagicMock()
+    sys.modules["envoy.service.ext_proc.v3"] = MagicMock()
+    sys.modules["envoy.service.ext_proc.v3.external_processor_pb2"] = mock_ep
+    sys.modules["envoy.service.ext_proc.v3.external_processor_pb2_grpc"] = (
+        mock_ep_grpc
+    )
+    sys.modules["envoy.config"] = MagicMock()
+    sys.modules["envoy.config.core"] = MagicMock()
+    sys.modules["envoy.config.core.v3"] = MagicMock()
+    sys.modules["envoy.config.core.v3.base_pb2"] = mock_core
+    sys.modules["envoy.type"] = MagicMock()
+    sys.modules["envoy.type.v3"] = MagicMock()
+    sys.modules["envoy.type.v3.http_status_pb2"] = mock_http_status
+
+    yield {
+        "ep": mock_ep,
+        "ep_grpc": mock_ep_grpc,
+        "core": mock_core,
+        "http_status": mock_http_status,
+    }
+
+    # Cleanup
+    for key in list(sys.modules.keys()):
+        if key.startswith("envoy"):
+            del sys.modules[key]
+    if "src.server" in sys.modules:
+        del sys.modules["src.server"]
+
+
+@pytest.fixture
+def mock_manager():
+    """Create a mock PluginManager."""
+    mock = Mock()
+    mock.invoke_hook = AsyncMock()
+    return mock
+
+
+@pytest.fixture
+def sample_tool_result_body():
+    """Create a sample tool result body."""
+    return {
+        "jsonrpc": "2.0",
+        "id": "test-123",
+        "result": {
+            "content": [{"type": "text", "text": "Tool execution result"}]
+        },
+    }
+
+
+@pytest.mark.asyncio
+async def test_getToolPostInvokeResponse_continue_processing(
+    mock_envoy_modules, mock_manager, sample_tool_result_body
+):
+    """Test getToolPostInvokeResponse when plugin allows processing to continue."""
+    # Setup mock response objects
+    mock_response = MagicMock()
+    mock_response.HasField.return_value = True
+    mock_response.response_body.response.HasField.return_value = False
+    mock_envoy_modules["ep"].ProcessingResponse.return_value = mock_response
+
+    # Import server after mocking
+    import src.server
+
+    # Setup mock to return continue_processing=True
+    mock_result = ToolPostInvokeResult(
+        continue_processing=True,
+        modified_payload=None,
+    )
+    mock_manager.invoke_hook.return_value = (mock_result, None)
+
+    # Inject mock manager
+    src.server.manager = mock_manager
+
+    # Call the function
+    _ = await src.server.getToolPostInvokeResponse(sample_tool_result_body)
+
+    # Verify the hook was called
+    assert mock_manager.invoke_hook.called
+    call_args = mock_manager.invoke_hook.call_args[0]
+    payload = call_args[1]
+    assert isinstance(payload, ToolPostInvokePayload)
+    assert payload.result == sample_tool_result_body["result"]
+    # assert payload.name == "replaceme" # Replace this after better naming
+
+
+@pytest.mark.asyncio
+async def test_getToolPostInvokeResponse_blocked(
+    mock_envoy_modules, mock_manager, sample_tool_result_body
+):
+    """Test getToolPostInvokeResponse when plugin blocks the response."""
+    # Import server after mocking
+    import src.server
+
+    # Setup mock to return continue_processing=False with violation
+    violation = PluginViolation(
+        reason="Sensitive content detected",
+        description="Tool response contains forbidden content",
+        code="CONTENT_VIOLATION",
+    )
+    mock_result = ToolPostInvokeResult(
+        continue_processing=False,
+        violation=violation,
+    )
+    mock_manager.invoke_hook.return_value = (mock_result, None)
+
+    # Inject mock manager
+    src.server.manager = mock_manager
+
+    # Call the function
+    response = await src.server.getToolPostInvokeResponse(
+        sample_tool_result_body
+    )
+
+    # Verify the hook was called with correct payload
+    assert mock_manager.invoke_hook.called
+    call_args = mock_manager.invoke_hook.call_args[0]
+    payload = call_args[1]
+    assert isinstance(payload, ToolPostInvokePayload)
+    assert payload.result == sample_tool_result_body["result"]
+
+    # Verify response was created (error path taken)
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_getToolPostInvokeResponse_modified_payload(
+    mock_envoy_modules, mock_manager, sample_tool_result_body
+):
+    """Test getToolPostInvokeResponse when plugin modifies the payload."""
+    # Import server after mocking
+    import src.server
+
+    # Setup mock to return modified payload
+    modified_result = {
+        "content": [{"type": "text", "text": "Modified tool result"}]
+    }
+    modified_payload = ToolPostInvokePayload(
+        name="test_tool", result=modified_result
+    )
+    mock_result = ToolPostInvokeResult(
+        continue_processing=True,
+        modified_payload=modified_payload,
+    )
+    mock_manager.invoke_hook.return_value = (mock_result, None)
+
+    # Inject mock manager
+    src.server.manager = mock_manager
+
+    # Call the function
+    response = await src.server.getToolPostInvokeResponse(
+        sample_tool_result_body
+    )
+
+    # Verify the hook was called
+    assert mock_manager.invoke_hook.called
+
+    # Verify response was created
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_getToolPostInvokeResponse_multiple_content_items(
+    mock_envoy_modules, mock_manager
+):
+    """Test getToolPostInvokeResponse with multiple content items."""
+    # Setup mock response
+    mock_response = MagicMock()
+    mock_envoy_modules["ep"].ProcessingResponse.return_value = mock_response
+
+    # Import server after mocking
+    import src.server
+
+    body = {
+        "jsonrpc": "2.0",
+        "id": "test-789",
+        "result": {
+            "content": [
+                {"type": "text", "text": "First item"},
+                {"type": "text", "text": "Second item"},
+                {"type": "image", "url": "http://example.com/img.png"},
+            ]
+        },
+    }
+
+    mock_result = ToolPostInvokeResult(continue_processing=True)
+    mock_manager.invoke_hook.return_value = (mock_result, None)
+
+    # Inject mock manager
+    src.server.manager = mock_manager
+
+    # Call the function
+    _ = await src.server.getToolPostInvokeResponse(body)
+
+    # Verify the payload passed to the hook contains all content
+    call_args = mock_manager.invoke_hook.call_args[0]
+    payload = call_args[1]
+    assert len(payload.result["content"]) == 3
+    assert payload.result["content"][0]["text"] == "First item"
+    assert payload.result["content"][1]["text"] == "Second item"
+    assert payload.result["content"][2]["url"] == "http://example.com/img.png"

From eafe4c2c4304bb18ca2fc2e7632452020862cfdd Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Thu, 12 Feb 2026 14:37:30 -0700
Subject: [PATCH 10/27] :recycle::white_check_mark: Tool response body
 buffering

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py        | 172 +++++++++++++++++++-------------------
 tests/test_server.py | 191 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 274 insertions(+), 89 deletions(-)

diff --git a/src/server.py b/src/server.py
index 4dac2aa..c5f209a 100644
--- a/src/server.py
+++ b/src/server.py
@@ -229,6 +229,85 @@ async def getPromptPreFetchResponse(body):
     return body_resp
 
 
+# ============================================================================
+# RESPONSE BODY PROCESSING HELPER
+# ============================================================================
+
+
+async def process_response_body_buffer(buffer: bytearray):
+    """Process buffered response body content.
+
+    Parses the buffered content (supporting both SSE and plain JSON-RPC formats),
+    and invokes the tool post-invoke hook if it's a tool result.
+
+    Args:
+        buffer: The accumulated response body bytes
+
+    Returns:
+        ProcessingResponse to send back to Envoy
+    """
+    if not buffer:
+        # Empty buffer at end of stream
+        logger.debug("End of stream with empty buffer")
+        return ep.ProcessingResponse(
+            response_body=ep.BodyResponse(response=ep.CommonResponse())
+        )
+
+    try:
+        text = buffer.decode("utf-8")
+    except UnicodeDecodeError:
+        logger.debug("Response body not UTF-8; skipping")
+        return ep.ProcessingResponse(
+            response_body=ep.BodyResponse(response=ep.CommonResponse())
+        )
+
+    lines = text.split("\n")
+    logger.debug(f"Response body text: {lines}")
+
+    # Handle both SSE format and plain JSON-RPC format
+    data = None
+
+    # Check if this is SSE format (starts with "event:" or "data:")
+    if text.strip().startswith(("event:", "data:")):
+        # Parse SSE format
+        lines = text.split("\n")
+        for line in lines:
+            line = line.strip()
+            if line.startswith("data:"):
+                json_str = line[5:].strip()  # Remove "data:" prefix
+                logger.debug(f"Extracted JSON from SSE: {json_str}")
+                try:
+                    data = json.loads(json_str)
+                    break
+                except json.JSONDecodeError:
+                    continue
+    else:
+        # Parse plain JSON-RPC format
+        lines = [line.strip() for line in text.split("\n") if line.strip()]
+        if lines:
+            try:
+                data = json.loads(lines[0])
+            except json.JSONDecodeError as e:
+                logger.error(f"Failed to parse JSON: {e}")
+
+    if data:
+        logger.debug(f"Parsed response data: {data}")
+
+        # Check if this is a tool result response
+        if "result" in data and "content" in data["result"]:
+            logger.info("Invoking tool post-invoke hook")
+            return await getToolPostInvokeResponse(data)
+        else:
+            return ep.ProcessingResponse(
+                response_body=ep.BodyResponse(response=ep.CommonResponse())
+            )
+    else:
+        logger.warning("No data parsed from response body")
+        return ep.ProcessingResponse(
+            response_body=ep.BodyResponse(response=ep.CommonResponse())
+        )
+
+
 # ============================================================================
 # ENVOY EXTERNAL PROCESSOR SERVICER
 # ============================================================================
@@ -356,95 +435,10 @@ async def Process(
                         "End of stream reached, processing complete buffered response"
                     )
 
-                    # Process buffered content if any
-                    if resp_body_buf:
-                        try:
-                            text = resp_body_buf.decode("utf-8")
-                        except UnicodeDecodeError:
-                            logger.debug("Response body not UTF-8; skipping")
-                            body_resp = ep.ProcessingResponse(
-                                response_body=ep.BodyResponse(
-                                    response=ep.CommonResponse()
-                                )
-                            )
-                        else:
-                            lines = text.split("\n")
-                            logger.debug(f"Response body text: {lines}")
-
-                            # Handle both SSE format and plain JSON-RPC format
-                            data = None
-
-                            # Check if this is SSE format (starts with "event:" or "data:")
-                            if text.strip().startswith(("event:", "data:")):
-                                # Parse SSE format
-                                lines = text.split("\n")
-                                for line in lines:
-                                    line = line.strip()
-                                    if line.startswith("data:"):
-                                        json_str = line[
-                                            5:
-                                        ].strip()  # Remove "data:" prefix
-                                        logger.debug(
-                                            f"Extracted JSON from SSE: {json_str}"
-                                        )
-                                        try:
-                                            data = json.loads(json_str)
-                                            break
-                                        except json.JSONDecodeError:
-                                            continue
-                            else:
-                                # Parse plain JSON-RPC format
-                                lines = [
-                                    line.strip()
-                                    for line in text.split("\n")
-                                    if line.strip()
-                                ]
-                                if lines:
-                                    try:
-                                        data = json.loads(lines[0])
-                                    except json.JSONDecodeError as e:
-                                        logger.error(
-                                            f"Failed to parse JSON: {e}"
-                                        )
-
-                            if data:
-                                logger.debug(f"Parsed response data: {data}")
-
-                                # Check if this is a tool result response
-                                if (
-                                    "result" in data
-                                    and "content" in data["result"]
-                                ):
-                                    logger.info(
-                                        "Invoking tool post-invoke hook"
-                                    )
-                                    body_resp = await getToolPostInvokeResponse(
-                                        data
-                                    )
-                                else:
-                                    body_resp = ep.ProcessingResponse(
-                                        response_body=ep.BodyResponse(
-                                            response=ep.CommonResponse()
-                                        )
-                                    )
-                            else:
-                                logger.warning(
-                                    "No data parsed from response body"
-                                )
-                                body_resp = ep.ProcessingResponse(
-                                    response_body=ep.BodyResponse(
-                                        response=ep.CommonResponse()
-                                    )
-                                )
-                    else:
-                        # Empty buffer at end of stream
-                        logger.debug("End of stream with empty buffer")
-                        body_resp = ep.ProcessingResponse(
-                            response_body=ep.BodyResponse(
-                                response=ep.CommonResponse()
-                            )
-                        )
-
+                    # Process the buffered content
+                    body_resp = await process_response_body_buffer(
+                        resp_body_buf
+                    )
                     yield body_resp
                     resp_body_buf.clear()
                 else:
diff --git a/tests/test_server.py b/tests/test_server.py
index 0a5f033..efc0fe7 100644
--- a/tests/test_server.py
+++ b/tests/test_server.py
@@ -6,6 +6,7 @@
 # Standard
 from unittest.mock import AsyncMock, Mock, MagicMock
 import sys
+import json
 
 # Third-Party
 import pytest
@@ -79,6 +80,33 @@ def sample_tool_result_body():
     }
 
 
+def setup_response_mocks(mock_envoy_modules):
+    """Setup common response mocks."""
+    mock_envoy_modules["ep"].ProcessingResponse.return_value = MagicMock()
+    mock_envoy_modules["ep"].BodyResponse.return_value = MagicMock()
+    mock_envoy_modules["ep"].CommonResponse.return_value = MagicMock()
+
+
+def setup_manager_with_result(mock_manager, continue_processing=True):
+    """Setup mock manager with a tool post-invoke result."""
+    mock_result = ToolPostInvokeResult(continue_processing=continue_processing)
+    mock_manager.invoke_hook.return_value = (mock_result, None)
+    return mock_manager
+
+
+def verify_payload_content(payload, expected_result, expected_text):
+    """Verify payload contains expected content."""
+    assert isinstance(payload, ToolPostInvokePayload)
+    assert payload.result == expected_result
+    assert payload.result["content"][0]["type"] == "text"
+    assert payload.result["content"][0]["text"] == expected_text
+
+
+# ============================================================================
+# Tool Post-Invoke Hook Tests
+# ============================================================================
+
+
 @pytest.mark.asyncio
 async def test_getToolPostInvokeResponse_continue_processing(
     mock_envoy_modules, mock_manager, sample_tool_result_body
@@ -230,3 +258,166 @@ async def test_getToolPostInvokeResponse_multiple_content_items(
     assert payload.result["content"][0]["text"] == "First item"
     assert payload.result["content"][1]["text"] == "Second item"
     assert payload.result["content"][2]["url"] == "http://example.com/img.png"
+
+
+# ============================================================================
+# Response Body Processing Tests
+# ============================================================================
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_with_tool_result(
+    mock_envoy_modules, mock_manager
+):
+    """Test process_response_body_buffer with a tool result."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    setup_manager_with_result(mock_manager)
+    src.server.manager = mock_manager
+
+    tool_result = {
+        "jsonrpc": "2.0",
+        "id": "test-123",
+        "result": {"content": [{"type": "text", "text": "Result"}]},
+    }
+    buffer = bytearray(json.dumps(tool_result).encode("utf-8"))
+    response = await src.server.process_response_body_buffer(buffer)
+
+    assert mock_manager.invoke_hook.called
+    payload = mock_manager.invoke_hook.call_args[0][1]
+    verify_payload_content(payload, tool_result["result"], "Result")
+    # Verify ProcessingResponse was returned
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_with_sse_format(
+    mock_envoy_modules, mock_manager
+):
+    """Test process_response_body_buffer with SSE formatted content."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    setup_manager_with_result(mock_manager)
+    src.server.manager = mock_manager
+
+    tool_result = {
+        "jsonrpc": "2.0",
+        "id": "test-sse",
+        "result": {"content": [{"type": "text", "text": "SSE data"}]},
+    }
+    sse_body = f"event: message\ndata: {json.dumps(tool_result)}\n\n"
+    buffer = bytearray(sse_body.encode("utf-8"))
+    response = await src.server.process_response_body_buffer(buffer)
+
+    assert mock_manager.invoke_hook.called
+    payload = mock_manager.invoke_hook.call_args[0][1]
+    verify_payload_content(payload, tool_result["result"], "SSE data")
+    # Verify ProcessingResponse was returned
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_empty(
+    mock_envoy_modules, mock_manager
+):
+    """Test process_response_body_buffer with empty buffer."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    src.server.manager = mock_manager
+    response = await src.server.process_response_body_buffer(bytearray())
+
+    # Verify ProcessingResponse was returned
+    assert response is not None
+    assert not mock_manager.invoke_hook.called, (
+        "Tool post-invoke hook should not be called for empty buffer"
+    )
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_non_tool_result(
+    mock_envoy_modules, mock_manager
+):
+    """Test process_response_body_buffer with non-tool result (error response)."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    src.server.manager = mock_manager
+
+    error_response = {
+        "jsonrpc": "2.0",
+        "id": "test-error",
+        "error": {"code": -32000, "message": "Error"},
+    }
+    buffer = bytearray(json.dumps(error_response).encode("utf-8"))
+    response = await src.server.process_response_body_buffer(buffer)
+
+    # Verify ProcessingResponse was returned
+    assert response is not None
+    assert not mock_manager.invoke_hook.called, (
+        "Tool post-invoke hook should not be called for error responses"
+    )
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_multiple_chunks_scenario(
+    mock_envoy_modules, mock_manager
+):
+    """Test buffering: content in chunks, then empty end_of_stream chunk.
+
+    Simulates: chunk1 (content) + chunk2 (content) + chunk3 (empty, end_of_stream).
+    """
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    setup_manager_with_result(mock_manager)
+    src.server.manager = mock_manager
+
+    tool_result = {
+        "jsonrpc": "2.0",
+        "id": "test-multi-chunk",
+        "result": {"content": [{"type": "text", "text": "Multi chunk data"}]},
+    }
+    body_bytes = json.dumps(tool_result).encode("utf-8")
+
+    # Simulate buffering: chunk1 + chunk2 + empty chunk
+    buffer = bytearray()
+    buffer.extend(body_bytes[:25])  # Chunk 1
+    buffer.extend(body_bytes[25:])  # Chunk 2
+    buffer.extend(b"")  # Chunk 3 (empty, triggers processing)
+
+    response = await src.server.process_response_body_buffer(buffer)
+
+    assert mock_manager.invoke_hook.called
+    payload = mock_manager.invoke_hook.call_args[0][1]
+    verify_payload_content(payload, tool_result["result"], "Multi chunk data")
+    # Verify ProcessingResponse was returned
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_single_chunk_with_end_of_stream(
+    mock_envoy_modules, mock_manager
+):
+    """Test buffering: all content in one chunk with end_of_stream."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    setup_manager_with_result(mock_manager)
+    src.server.manager = mock_manager
+
+    tool_result = {
+        "jsonrpc": "2.0",
+        "id": "test-single",
+        "result": {"content": [{"type": "text", "text": "Single chunk"}]},
+    }
+    buffer = bytearray(json.dumps(tool_result).encode("utf-8"))
+    response = await src.server.process_response_body_buffer(buffer)
+
+    assert mock_manager.invoke_hook.called
+    payload = mock_manager.invoke_hook.call_args[0][1]
+    verify_payload_content(payload, tool_result["result"], "Single chunk")
+    # Verify ProcessingResponse was returned
+    assert response is not None

From 6246685fb9aa1cc84e38600944e166d3605942ed Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 12:32:26 -0700
Subject: [PATCH 11/27] :recycle: First pass nemocheck as internal and
 nemocheck_external as external plugin

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/README.md          | 137 +++++------------
 .../nemocheck/{nemocheck => }/__init__.py     |   0
 .../config.yaml                               |   4 +-
 .../nemocheck/nemocheck/plugin-manifest.yaml  |   9 --
 .../examples/nemocheck/nemocheck/plugin.py    | 145 ------------------
 .../plugin-manifest.yaml                      |   2 +-
 .../plugin.py                                 | 108 +++++++------
 plugins/examples/nemocheck/tests/__init__.py  |   1 +
 .../nemocheck/tests/test_nemocheck.py         |   8 +-
 .../.dockerignore                             |   0
 .../.env.template                             |   0
 .../.ruff.toml                                |   0
 .../Containerfile                             |   0
 .../MANIFEST.in                               |   0
 .../Makefile                                  |   0
 plugins/examples/nemocheck_external/README.md | 123 +++++++++++++++
 .../k8deploy/Makefile                         |   0
 .../k8deploy/config-tools.yaml                |   0
 .../k8deploy/deploy.yaml                      |   0
 .../k8deploy/server.yaml                      |   0
 .../pyproject.toml                            |   0
 .../resources/plugins/config.yaml             |   4 +-
 .../run-server.sh                             |   0
 .../{nemocheck => nemocheck_external}/uv.lock |   0
 plugins/examples/nemocheckinternal/README.md  |  42 -----
 .../examples/nemocheckinternal/__init__.py    |   7 -
 26 files changed, 229 insertions(+), 361 deletions(-)
 rename plugins/examples/nemocheck/{nemocheck => }/__init__.py (100%)
 rename plugins/examples/{nemocheckinternal => nemocheck}/config.yaml (95%)
 delete mode 100644 plugins/examples/nemocheck/nemocheck/plugin-manifest.yaml
 delete mode 100644 plugins/examples/nemocheck/nemocheck/plugin.py
 rename plugins/examples/{nemocheckinternal => nemocheck}/plugin-manifest.yaml (91%)
 rename plugins/examples/{nemocheckinternal => nemocheck}/plugin.py (59%)
 rename plugins/examples/{nemocheck => nemocheck_external}/.dockerignore (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/.env.template (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/.ruff.toml (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/Containerfile (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/MANIFEST.in (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/Makefile (100%)
 create mode 100644 plugins/examples/nemocheck_external/README.md
 rename plugins/examples/{nemocheck => nemocheck_external}/k8deploy/Makefile (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/k8deploy/config-tools.yaml (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/k8deploy/deploy.yaml (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/k8deploy/server.yaml (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/pyproject.toml (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/resources/plugins/config.yaml (93%)
 rename plugins/examples/{nemocheck => nemocheck_external}/run-server.sh (100%)
 rename plugins/examples/{nemocheck => nemocheck_external}/uv.lock (100%)
 delete mode 100644 plugins/examples/nemocheckinternal/README.md
 delete mode 100644 plugins/examples/nemocheckinternal/__init__.py

diff --git a/plugins/examples/nemocheck/README.md b/plugins/examples/nemocheck/README.md
index 48d3570..eb064ef 100644
--- a/plugins/examples/nemocheck/README.md
+++ b/plugins/examples/nemocheck/README.md
@@ -1,94 +1,44 @@
-# NemoCheck for Plugin Adapter
+# NemoCheck Internal Plugin
 
-Adapter for Nemo-Check guardrails.
+This directory contains the core `NemoCheck` plugin implementation used by both internal and external plugins.
 
-
-## Run plugin in kind cluster
-
- 1. Run Nemo Guardrails check server. Instructions [here](#Deploy-checkserver)
- 1. Update `CHECK_ENDPOINT` variable in k8deploy/deploy.yaml to point to guardrails check server endpoint
-
-    ```bash
-    cd plugins-adapter/plugins/examples/nemocheck
-    make deploy
-    ```
- 1.
-    <details>
-    <summary>Non-kind k8 cluster instructions</summary>
-
-    ```bash
-        cd plugins-adapter/plugins/examples/nemocheck
-        make container-build
-        # push image to your container repo and update image name in k8deploy/deploy.yaml
-        kubectl apply -f k8deploy/deploy.yaml
-
-    ```
-    </details>
-
- 1. Update plugin adapter to call this as an external plugin
-   
-   ```bash
-   cd ../../.. #project root directory plugins-adapter`
-   cp resources/config/external_plugin_nemocheck.yaml resources/config/config.yaml 
-   make all
-   ```
-
-## Test with MCP inspector
- * Add allowed tools to `plugins-adapter/plugins/examples/nemocheck/k8deploy/config-tools.yaml#check_tool_call_safety` 
-<table>
-<tr>
-<th> config-tools.yaml line-127</th>
-<th>Updated to add test2_hello_world </th>
-</tr>
-<tr>
-<td>
-<pre>
-
-```python
-@action(is_system_action=True)
-async def check_tool_call_safety(tool_calls=None, context=None):
-    """Allow list for tool execution."""
-      ...
-      allowed_tools = ["get_weather", "search_web", 
-          "get_time", "slack_read_messages"]
-      ...
-```  
-</pre>
-</td>
-<td>
-
-```python
-@action(is_system_action=True)
-async def check_tool_call_safety(tool_calls=None, context=None):
-    """Allow list for tool execution."""
-      ...
-      allowed_tools = ["get_weather", "search_web", "get_time", 
-          "test2_hello_world", "slack_read_messages"]
-      ...
-```  
-
-</td>
-</tr>
-</table>
-
-
- * Redeploy check server
- * Open mcp inspector. Try tools in allow list vs tools not in allow list
-
-
-## Deploy-checkserver
+## Prerequisites: Nemo-check server
  * Refer to [orignal repo](https://github.com/m-misiura/demos/tree/main/nemo_openshift/guardrail-checks/deployment) for full instructions
  * Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
- * Makefile has targets to load checkserver to kind cluster, etc.
 
    ```bash
-   cd plugins-adapter/plugins/examples/nemocheck/k8deploy
-   make deploy
+	docker pull quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1
+	kind load docker-image quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 --name mcp-gateway
+    cd plugins-adapter/plugins/examples/nemocheck/k8deploy
+	kubectl apply -f config-tools.yaml
+	kubectl apply -f server.yaml
 
    ```
+## Installation
+
+1. Find url of nemo-check-server service. E.g., from svc in `server.yaml`
+1. Update `${project_root}/resources/config/config.yaml`. Add the blob below, merge if other `plugin`s or `plugin_dir`s already exists. Sample file [here](/resources/config/nemocheck-internal-config.yaml)
+
+    ```yaml
+    # plugins/config.yaml - Main plugin configuration file
+    plugins:
+      - name: "NemoCheck"
+        kind: "plugins.examples.nemocheck.nemocheck.plugin.NemoCheck"
+        description: "Adapter for nemo check server"
+        version: "0.1.0"
+        hooks: ["tool_pre_invoke", "tool_post_invoke"]
+        mode: "enforce"  # enforce | permissive | disabled
+        config:
+          checkserver_url: "http://nemo-guardrails-service:8000/v1/guardrail/checks"
+    # Plugin directories to scan
+    plugin_dirs:
+      - "plugins/examples/nemocheck"    # Nemo Check Server plugins
+    ```
 
+1. In `config.yaml` ensure key `plugins.config.checkserver_url` points to the correct service
+1. Start plugin adapter
 
-## Plugin Development 
+## Plugin Development
 
 To install dependencies with dev packages (required for linting and testing):
 
@@ -121,30 +71,13 @@ make test
 
 ## Code Linting
 
-Before checking in any code for the project, please lint the code.  This can be done using:
+Before checking in any code for the project, please lint the code. This can be done using:
 
 ```bash
 make lint-fix
 ```
 
-## Runtime (server)
+# Test
 
-This project uses [chuck-mcp-runtime](https://github.com/chrishayuk/chuk-mcp-runtime) to run external plugins as a standardized MCP server.
-
-To build the container image:
-
-```bash
-make build
-```
-
-To run the container:
-
-```bash
-make start
-```
-
-To stop the container:
-
-```bash
-make stop
-```
+1. Open mcp-inspector to the mcp-gateway
+1. Try running a tool configured/not configured in nemo check config allow list in configmap [E.g.](/plugins/examples/nemocheck/k8deploy/config-tools.yaml)
diff --git a/plugins/examples/nemocheck/nemocheck/__init__.py b/plugins/examples/nemocheck/__init__.py
similarity index 100%
rename from plugins/examples/nemocheck/nemocheck/__init__.py
rename to plugins/examples/nemocheck/__init__.py
diff --git a/plugins/examples/nemocheckinternal/config.yaml b/plugins/examples/nemocheck/config.yaml
similarity index 95%
rename from plugins/examples/nemocheckinternal/config.yaml
rename to plugins/examples/nemocheck/config.yaml
index d930000..d9f97b1 100644
--- a/plugins/examples/nemocheckinternal/config.yaml
+++ b/plugins/examples/nemocheck/config.yaml
@@ -1,5 +1,5 @@
 plugins:
-  - name: "NemoCheckv2"
+  - name: "NemoCheck"
     kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2"
     description: "Nemo Check Adapter"
     version: "0.1.0"
@@ -17,7 +17,7 @@ plugins:
 
 # Plugin directories to scan
 plugin_dirs:
-  - "nemocheckv2"
+  - "nemocheck"
 
 # Global plugin settings
 plugin_settings:
diff --git a/plugins/examples/nemocheck/nemocheck/plugin-manifest.yaml b/plugins/examples/nemocheck/nemocheck/plugin-manifest.yaml
deleted file mode 100644
index 3e8fe2e..0000000
--- a/plugins/examples/nemocheck/nemocheck/plugin-manifest.yaml
+++ /dev/null
@@ -1,9 +0,0 @@
-description: "Adapter for Nemo-Check guardrails"
-author: "julianstephen"
-version: "0.1.0"
-available_hooks:
-  - "prompt_pre_hook"
-  - "prompt_post_hook"
-  - "tool_pre_hook"
-  - "tool_post_hook"
-default_configs:
diff --git a/plugins/examples/nemocheck/nemocheck/plugin.py b/plugins/examples/nemocheck/nemocheck/plugin.py
deleted file mode 100644
index c5edce8..0000000
--- a/plugins/examples/nemocheck/nemocheck/plugin.py
+++ /dev/null
@@ -1,145 +0,0 @@
-"""Adapter for Nemo-Check guardrails.
-
-Copyright 2025
-SPDX-License-Identifier: Apache-2.0
-Authors: julianstephen
-
-This module loads configurations for plugins.
-"""
-
-# First-Party
-from mcpgateway.plugins.framework import (
-    Plugin,
-    PluginConfig,
-    PluginContext,
-    PromptPosthookPayload,
-    PromptPosthookResult,
-    PromptPrehookPayload,
-    PromptPrehookResult,
-    ToolPostInvokePayload,
-    ToolPostInvokeResult,
-    ToolPreInvokePayload,
-    ToolPreInvokeResult,
-    PluginViolation,
-)
-
-
-import logging
-import os
-import requests
-
-headers = {
-    "Content-Type": "application/json",
-}
-# Initialize logging service first
-logger = logging.getLogger(__name__)
-log_level = os.getenv("LOGLEVEL", "INFO").upper()
-logger.setLevel(log_level)
-
-MODEL_NAME = os.getenv("NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct")  # Currently only for logging.
-CHECK_ENDPOINT = os.getenv("CHECK_ENDPOINT", "http://nemo-guardrails-service:8000")
-
-
-class NemoCheck(Plugin):
-    """Adapter for Nemo-Check guardrails."""
-
-    def __init__(self, config: PluginConfig):
-        """Entry init block for plugin.
-
-        Args:
-          logger: logger that the skill can make use of
-          config: the skill configuration
-        """
-        super().__init__(config)
-
-    async def prompt_pre_fetch(self, payload: PromptPrehookPayload, context: PluginContext) -> PromptPrehookResult:
-        """The plugin hook run before a prompt is retrieved and rendered.
-
-        Args:
-            payload: The prompt payload to be analyzed.
-            context: contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the prompt can proceed.
-        """
-
-        return PromptPrehookResult(continue_processing=True)
-
-    async def prompt_post_fetch(self, payload: PromptPosthookPayload, context: PluginContext) -> PromptPosthookResult:
-        """Plugin hook run after a prompt is rendered.
-
-        Args:
-            payload: The prompt payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the prompt can proceed.
-        """
-        return PromptPosthookResult(continue_processing=True)
-
-    async def tool_pre_invoke(self, payload: ToolPreInvokePayload, context: PluginContext) -> ToolPreInvokeResult:
-        """Plugin hook run before a tool is invoked.
-
-        Args:
-            payload: The tool payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the tool can proceed.
-        """
-        logger.info("tool_pre_invoke....")
-        logger.info(payload)
-        tool_name = payload.name  # ("tool_name", None)
-        check_nemo_payload = {
-            "model": MODEL_NAME,
-            "messages": [
-                {
-                    "role": "assistant",
-                    "tool_calls": [
-                        {
-                            "id": "call_plug_adap_nem_check_123",
-                            "type": "function",
-                            "function": {
-                                "name": tool_name,
-                                "arguments": payload.args.get("tool_args", None),
-                            },
-                        }
-                    ],
-                }
-            ],
-        }
-        violation = None
-        response = requests.post(CHECK_ENDPOINT, headers=headers, json=check_nemo_payload)
-        if response.status_code == 200:
-            data = response.json()
-            status = data.get("status", "blocked")
-            logger.debug(f"rails reply:{data}")
-            if status == "success":
-                metadata = data.get("rails_status")
-                result = ToolPreInvokeResult(continue_processing=True, metadata=metadata)
-            else:
-                metadata = data.get("rails_status")
-                violation = PluginViolation(
-                    reason=f"Tool Check status:{status}", description="Rails check blocked request", code=f"checkserver_http_status_code:{response.status_code}", details=metadata
-                )
-                result = ToolPreInvokeResult(continue_processing=False, violation=violation, metadata=metadata)
-
-        else:
-            violation = PluginViolation(
-                reason="Tool Check Unavailable", description="Tool arguments check server returned error:", code=f"checkserver_http_status_code:{response.status_code}", details={}
-            )
-            result = ToolPreInvokeResult(continue_processing=False, violation=violation)
-
-        return result
-
-    async def tool_post_invoke(self, payload: ToolPostInvokePayload, context: PluginContext) -> ToolPostInvokeResult:
-        """Plugin hook run after a tool is invoked.
-
-        Args:
-            payload: The tool result payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the tool result should proceed.
-        """
-        return ToolPostInvokeResult(continue_processing=True)
diff --git a/plugins/examples/nemocheckinternal/plugin-manifest.yaml b/plugins/examples/nemocheck/plugin-manifest.yaml
similarity index 91%
rename from plugins/examples/nemocheckinternal/plugin-manifest.yaml
rename to plugins/examples/nemocheck/plugin-manifest.yaml
index d8fa01f..779bf7b 100644
--- a/plugins/examples/nemocheckinternal/plugin-manifest.yaml
+++ b/plugins/examples/nemocheck/plugin-manifest.yaml
@@ -1,5 +1,5 @@
 description: "Nemo Check Adapter"
-name: NemoCheckv2
+name: NemoCheck
 author: "julianstephen"
 version: "0.1.0"
 available_hooks:
diff --git a/plugins/examples/nemocheckinternal/plugin.py b/plugins/examples/nemocheck/plugin.py
similarity index 59%
rename from plugins/examples/nemocheckinternal/plugin.py
rename to plugins/examples/nemocheck/plugin.py
index 30b3223..598bb4e 100644
--- a/plugins/examples/nemocheckinternal/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -1,10 +1,10 @@
-"""Nemo Check Adapter.
+"""Nemo Check Plugin
 
 Copyright 2025
 SPDX-License-Identifier: Apache-2.0
 Authors: julianstephen
 
-This module loads configurations for plugins.
+This module provides the core Nemo Check guardrails plugin implementation.
 """
 
 # First-Party
@@ -26,41 +26,37 @@
 import logging
 import os
 import requests
-import json
 
-# Initialize logging service first
+# Initialize logging
 logger = logging.getLogger(__name__)
 log_level = os.getenv("LOGLEVEL", "INFO").upper()
 logger.setLevel(log_level)
 
-MODEL_NAME = os.getenv(
-    "NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct"
-)  # Currently only for logging.
-CHECK_ENDPOINT = os.getenv("CHECK_ENDPOINT", "http://nemo-guardrails-service:8000")
-
+MODEL_NAME = os.getenv("NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct")
+DEFAULT_CHECK_ENDPOINT = os.getenv(
+    "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000"
+)
 
-headers = {
+HEADERS = {
     "Content-Type": "application/json",
 }
 
 
-class NemoCheckv2(Plugin):
-    """Nemo Check Adapter."""
+class NemoCheck(Plugin):
+    """Nemo Check guardrails plugin."""
 
     def __init__(self, config: PluginConfig):
-        """Entry init block for plugin.
+        """Initialize the plugin.
 
         Args:
-          logger: logger that the skill can make use of
-          config: the skill configuration
+            config: The plugin configuration
         """
-        global CHECK_ENDPOINT
-        logger.info(f"plugin config {config}")
-        endpoint = config.config.get("checkserver_url", None)
-        if endpoint is not None:
-            CHECK_ENDPOINT = endpoint
-        logger.info(f"checkserver at {config}:{CHECK_ENDPOINT}")
         super().__init__(config)
+        # Allow config to override the endpoint
+        self.check_endpoint = config.config.get(
+            "checkserver_url", DEFAULT_CHECK_ENDPOINT
+        )
+        logger.info(f"Nemo Check endpoint: {self.check_endpoint}")
 
     async def prompt_pre_fetch(
         self, payload: PromptPrehookPayload, context: PluginContext
@@ -69,10 +65,10 @@ async def prompt_pre_fetch(
 
         Args:
             payload: The prompt payload to be analyzed.
-            context: contextual information about the hook call.
+            context: Contextual information about the hook call.
 
         Returns:
-            The result of the plugin's analysis, including whether the prompt can proceed.
+            The result of the plugin's analysis.
         """
         return PromptPrehookResult(continue_processing=True)
 
@@ -123,41 +119,55 @@ async def tool_pre_invoke(
                 }
             ],
         }
-        violation = None
-        response = requests.post(
-            CHECK_ENDPOINT, headers=headers, json=check_nemo_payload
-        )
-        if response.status_code == 200:
-            data = response.json()
-            status = data.get("status", "blocked")
-            logger.debug(f"rails reply:{data}")
-            if status == "success":
-                metadata = data.get("rails_status")
-                result = ToolPreInvokeResult(
-                    continue_processing=True, metadata=metadata
-                )
+
+        try:
+            response = requests.post(
+                self.check_endpoint, headers=HEADERS, json=check_nemo_payload
+            )
+
+            if response.status_code == 200:
+                data = response.json()
+                status = data.get("status", "blocked")
+                logger.debug(f"rails reply: {data}")
+
+                if status == "success":
+                    metadata = data.get("rails_status")
+                    return ToolPreInvokeResult(
+                        continue_processing=True, metadata=metadata
+                    )
+                else:
+                    metadata = data.get("rails_status")
+                    violation = PluginViolation(
+                        reason=f"Tool Check status: {status}",
+                        description="Rails check blocked request",
+                        code=f"checkserver_http_status_code:{response.status_code}",
+                        details=metadata,
+                    )
+                    return ToolPreInvokeResult(
+                        continue_processing=False,
+                        violation=violation,
+                        metadata=metadata,
+                    )
             else:
-                metadata = data.get("rails_status")
                 violation = PluginViolation(
-                    reason=f"Check tool rails:{status}.",
-                    description=json.dumps(data),
+                    reason="Tool Check Unavailable",
+                    description="Tool arguments check server returned error",
                     code=f"checkserver_http_status_code:{response.status_code}",
-                    details=metadata,
+                    details={},
                 )
-                result = ToolPreInvokeResult(
-                    continue_processing=False, violation=violation, metadata=metadata
+                return ToolPreInvokeResult(
+                    continue_processing=False, violation=violation
                 )
 
-        else:
+        except Exception as e:
+            logger.error(f"Error calling Nemo Check endpoint: {e}")
             violation = PluginViolation(
-                reason="Tool Check Unavailable",
-                description="Tool arguments check server returned error:",
-                code=f"checkserver_http_status_code:{response.status_code}",
+                reason="Tool Check Error",
+                description=f"Failed to connect to check server: {str(e)}",
+                code="checkserver_connection_error",
                 details={},
             )
-            result = ToolPreInvokeResult(continue_processing=False, violation=violation)
-
-        return result
+            return ToolPreInvokeResult(continue_processing=False, violation=violation)
 
     async def tool_post_invoke(
         self, payload: ToolPostInvokePayload, context: PluginContext
diff --git a/plugins/examples/nemocheck/tests/__init__.py b/plugins/examples/nemocheck/tests/__init__.py
index e69de29..53386c4 100644
--- a/plugins/examples/nemocheck/tests/__init__.py
+++ b/plugins/examples/nemocheck/tests/__init__.py
@@ -0,0 +1 @@
+"""Tests for nemocheckinternal plugin."""
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 4321c51..8cd5561 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -4,13 +4,15 @@
 import pytest
 
 # First-Party
-from nemocheck.plugin import NemoCheck
 from mcpgateway.plugins.framework import (
     PluginConfig,
     GlobalContext,
     PromptPrehookPayload,
 )
 
+# Local
+from plugin import NemoCheck
+
 
 @pytest.mark.asyncio
 async def test_nemocheck():
@@ -25,7 +27,9 @@ async def test_nemocheck():
     plugin = NemoCheck(config)
 
     # Test your plugin logic
-    payload = PromptPrehookPayload(prompt_id="test_prompt", args={"arg0": "This is an argument"})
+    payload = PromptPrehookPayload(
+        prompt_id="test_prompt", args={"arg0": "This is an argument"}
+    )
     context = GlobalContext(request_id="1")
     result = await plugin.prompt_pre_fetch(payload, context)
     assert result.continue_processing
diff --git a/plugins/examples/nemocheck/.dockerignore b/plugins/examples/nemocheck_external/.dockerignore
similarity index 100%
rename from plugins/examples/nemocheck/.dockerignore
rename to plugins/examples/nemocheck_external/.dockerignore
diff --git a/plugins/examples/nemocheck/.env.template b/plugins/examples/nemocheck_external/.env.template
similarity index 100%
rename from plugins/examples/nemocheck/.env.template
rename to plugins/examples/nemocheck_external/.env.template
diff --git a/plugins/examples/nemocheck/.ruff.toml b/plugins/examples/nemocheck_external/.ruff.toml
similarity index 100%
rename from plugins/examples/nemocheck/.ruff.toml
rename to plugins/examples/nemocheck_external/.ruff.toml
diff --git a/plugins/examples/nemocheck/Containerfile b/plugins/examples/nemocheck_external/Containerfile
similarity index 100%
rename from plugins/examples/nemocheck/Containerfile
rename to plugins/examples/nemocheck_external/Containerfile
diff --git a/plugins/examples/nemocheck/MANIFEST.in b/plugins/examples/nemocheck_external/MANIFEST.in
similarity index 100%
rename from plugins/examples/nemocheck/MANIFEST.in
rename to plugins/examples/nemocheck_external/MANIFEST.in
diff --git a/plugins/examples/nemocheck/Makefile b/plugins/examples/nemocheck_external/Makefile
similarity index 100%
rename from plugins/examples/nemocheck/Makefile
rename to plugins/examples/nemocheck_external/Makefile
diff --git a/plugins/examples/nemocheck_external/README.md b/plugins/examples/nemocheck_external/README.md
new file mode 100644
index 0000000..53032ee
--- /dev/null
+++ b/plugins/examples/nemocheck_external/README.md
@@ -0,0 +1,123 @@
+# NemoCheck External Plugin
+
+This is an external plugin deployment for the NemoCheck guardrails adapter. It references the core `NemoCheck` implementation from the `nemocheck` plugin directory.
+
+## Architecture
+
+- **Core Implementation**: `plugins/examples/nemocheck/plugin.py` contains the actual NemoCheck logic
+- **External Deployment**: This directory (`nemocheck_external`) configures and deploys the plugin as an external MCP server
+- **Shared Logic**: Both internal and external deployments use the same underlying implementation to eliminate code duplication
+
+
+## Run plugin in kind cluster
+
+ 1. Run Nemo Guardrails check server. Instructions [here](#Deploy-checkserver)
+ 1. Update `CHECK_ENDPOINT` variable in k8deploy/deploy.yaml to point to guardrails check server endpoint
+
+    ```bash
+    cd plugins-adapter/plugins/examples/nemocheck
+    make deploy
+    ```
+ 1.
+    <details>
+    <summary>Non-kind k8 cluster instructions</summary>
+
+    ```bash
+        cd plugins-adapter/plugins/examples/nemocheck
+        make container-build
+        # push image to your container repo and update image name in k8deploy/deploy.yaml
+        kubectl apply -f k8deploy/deploy.yaml
+
+    ```
+    </details>
+
+ 1. Update plugin adapter to call this as an external plugin
+
+   ```bash
+   cd ../../.. #project root directory plugins-adapter`
+   cp resources/config/external_plugin_nemocheck.yaml resources/config/config.yaml
+   make all
+   ```
+
+## Test with MCP inspector
+ * Add allowed tools to `plugins-adapter/plugins/examples/nemocheck/k8deploy/config-tools.yaml#check_tool_call_safety`
+<table>
+<tr>
+<th> config-tools.yaml line-127</th>
+<th>Updated to add test2_hello_world </th>
+</tr>
+<tr>
+<td>
+<pre>
+
+```python
+@action(is_system_action=True)
+async def check_tool_call_safety(tool_calls=None, context=None):
+    """Allow list for tool execution."""
+      ...
+      allowed_tools = ["get_weather", "search_web",
+          "get_time", "slack_read_messages"]
+      ...
+```
+</pre>
+</td>
+<td>
+
+```python
+@action(is_system_action=True)
+async def check_tool_call_safety(tool_calls=None, context=None):
+    """Allow list for tool execution."""
+      ...
+      allowed_tools = ["get_weather", "search_web", "get_time",
+          "test2_hello_world", "slack_read_messages"]
+      ...
+```
+
+</td>
+</tr>
+</table>
+
+
+ * Redeploy check server
+ * Open mcp inspector. Try tools in allow list vs tools not in allow list
+
+
+## Deploy-checkserver
+ * Refer to [orignal repo](https://github.com/m-misiura/demos/tree/main/nemo_openshift/guardrail-checks/deployment) for full instructions
+ * Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
+ * Makefile has targets to load checkserver to kind cluster, etc.
+
+   ```bash
+   cd plugins-adapter/plugins/examples/nemocheck/k8deploy
+   make deploy
+
+   ```
+
+
+## Testing
+
+Tests are located in the `nemocheck` plugin directory since both deployments share the same core implementation.
+
+See the [nemocheck README](../nemocheck/README.md#testing) for testing instructions.
+
+## Runtime (server)
+
+This project uses [chuck-mcp-runtime](https://github.com/chrishayuk/chuk-mcp-runtime) to run external plugins as a standardized MCP server.
+
+To build the container image:
+
+```bash
+make build
+```
+
+To run the container:
+
+```bash
+make start
+```
+
+To stop the container:
+
+```bash
+make stop
+```
diff --git a/plugins/examples/nemocheck/k8deploy/Makefile b/plugins/examples/nemocheck_external/k8deploy/Makefile
similarity index 100%
rename from plugins/examples/nemocheck/k8deploy/Makefile
rename to plugins/examples/nemocheck_external/k8deploy/Makefile
diff --git a/plugins/examples/nemocheck/k8deploy/config-tools.yaml b/plugins/examples/nemocheck_external/k8deploy/config-tools.yaml
similarity index 100%
rename from plugins/examples/nemocheck/k8deploy/config-tools.yaml
rename to plugins/examples/nemocheck_external/k8deploy/config-tools.yaml
diff --git a/plugins/examples/nemocheck/k8deploy/deploy.yaml b/plugins/examples/nemocheck_external/k8deploy/deploy.yaml
similarity index 100%
rename from plugins/examples/nemocheck/k8deploy/deploy.yaml
rename to plugins/examples/nemocheck_external/k8deploy/deploy.yaml
diff --git a/plugins/examples/nemocheck/k8deploy/server.yaml b/plugins/examples/nemocheck_external/k8deploy/server.yaml
similarity index 100%
rename from plugins/examples/nemocheck/k8deploy/server.yaml
rename to plugins/examples/nemocheck_external/k8deploy/server.yaml
diff --git a/plugins/examples/nemocheck/pyproject.toml b/plugins/examples/nemocheck_external/pyproject.toml
similarity index 100%
rename from plugins/examples/nemocheck/pyproject.toml
rename to plugins/examples/nemocheck_external/pyproject.toml
diff --git a/plugins/examples/nemocheck/resources/plugins/config.yaml b/plugins/examples/nemocheck_external/resources/plugins/config.yaml
similarity index 93%
rename from plugins/examples/nemocheck/resources/plugins/config.yaml
rename to plugins/examples/nemocheck_external/resources/plugins/config.yaml
index 81d8b9f..bf3f0db 100644
--- a/plugins/examples/nemocheck/resources/plugins/config.yaml
+++ b/plugins/examples/nemocheck_external/resources/plugins/config.yaml
@@ -1,6 +1,6 @@
 plugins:
   - name: "NemoCheck"
-    kind: "nemocheck.plugin.NemoCheck"
+    kind: "plugin.NemoCheck"
     description: "Adapter for Nemo-Check guardrails"
     version: "0.1.0"
     author: "julianstephen"
@@ -17,7 +17,7 @@ plugins:
 
 # Plugin directories to scan
 plugin_dirs:
-  - "nemocheck"
+  - "../nemocheck"
 
 # Global plugin settings
 plugin_settings:
diff --git a/plugins/examples/nemocheck/run-server.sh b/plugins/examples/nemocheck_external/run-server.sh
similarity index 100%
rename from plugins/examples/nemocheck/run-server.sh
rename to plugins/examples/nemocheck_external/run-server.sh
diff --git a/plugins/examples/nemocheck/uv.lock b/plugins/examples/nemocheck_external/uv.lock
similarity index 100%
rename from plugins/examples/nemocheck/uv.lock
rename to plugins/examples/nemocheck_external/uv.lock
diff --git a/plugins/examples/nemocheckinternal/README.md b/plugins/examples/nemocheckinternal/README.md
deleted file mode 100644
index 0e874e2..0000000
--- a/plugins/examples/nemocheckinternal/README.md
+++ /dev/null
@@ -1,42 +0,0 @@
-# Internal NemoCheck Plugin
-
-## Prerequisites: Nemo-check server
- * Refer to [orignal repo](https://github.com/m-misiura/demos/tree/main/nemo_openshift/guardrail-checks/deployment) for full instructions
- * Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
- 
-   ```bash
-	docker pull quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1
-	kind load docker-image quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 --name mcp-gateway
-    cd plugins-adapter/plugins/examples/nemocheck/k8deploy
-	kubectl apply -f config-tools.yaml
-	kubectl apply -f server.yaml
-
-   ```
-## Installation
-
-1. Find url of nemo-check-server service. E.g., from svc in `server.yaml`
-1. Update `${project_root}/resources/config/config.yaml`. Add the blob below, merge if other `plugin`s or `plugin_dir`s already exists. Sample file [here](/resources/config/nemocheck-internal-config.yaml)
-
-    ```yaml
-    # plugins/config.yaml - Main plugin configuration file
-    plugins:
-      - name: "NemoCheckv2"
-        kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2"
-        description: "Adapter for nemo check server"
-        version: "0.1.0"        
-        hooks: ["tool_pre_invoke", "tool_post_invoke"]
-        mode: "enforce"  # enforce | permissive | disabled
-        config:
-          checkserver_url: "http://nemo-guardrails-service:8000/v1/guardrail/checks"    
-    # Plugin directories to scan
-    plugin_dirs:
-      - "plugins/examples/nemocheckinternal"    # Nemo Check Server plugins
-    ```
-
-1. In `config.yaml` ensure key `plugins.config.checkserver_url` points to the correct service
-1. Start plugin adapter
-
-# Test
-
-1. Open mcp-inspector to the mcp-gateway
-1. Try running a tool configured/not configured in nemo check config allow list in configmap [E.g.](/plugins/examples/nemocheck/k8deploy/config-tools.yaml)
\ No newline at end of file
diff --git a/plugins/examples/nemocheckinternal/__init__.py b/plugins/examples/nemocheckinternal/__init__.py
deleted file mode 100644
index 16e5be5..0000000
--- a/plugins/examples/nemocheckinternal/__init__.py
+++ /dev/null
@@ -1,7 +0,0 @@
-"""MCP Gateway NemoCheckv2 Plugin - Nemo Check Adapter.
-
-Copyright 2025
-SPDX-License-Identifier: Apache-2.0
-Authors: julianstephen
-
-"""

From 9d8eaa7a51518bf3e8bd593d09268a6225921274 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 13:10:52 -0700
Subject: [PATCH 12/27] :truck::recycle: Move around deploy files and
 references

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/README.md                    | 14 +--
 plugins/examples/nemocheck/README.md          | 87 ++++++++++---------
 plugins/examples/nemocheck/config.yaml        |  2 +-
 .../k8deploy/Makefile                         |  0
 .../k8deploy/config-tools.yaml                |  0
 .../k8deploy/server.yaml                      |  0
 plugins/examples/nemocheck/plugin.py          | 10 ++-
 plugins/examples/nemocheck/tests/__init__.py  |  2 +-
 plugins/examples/nemocheck_external/README.md | 71 +++------------
 resources/config/config.yaml                  | 17 ++--
 .../config/nemocheck-internal-config.yaml     | 14 +--
 11 files changed, 93 insertions(+), 124 deletions(-)
 rename plugins/examples/{nemocheck_external => nemocheck}/k8deploy/Makefile (100%)
 rename plugins/examples/{nemocheck_external => nemocheck}/k8deploy/config-tools.yaml (100%)
 rename plugins/examples/{nemocheck_external => nemocheck}/k8deploy/server.yaml (100%)

diff --git a/plugins/examples/README.md b/plugins/examples/README.md
index 7bc87b0..c857d33 100644
--- a/plugins/examples/README.md
+++ b/plugins/examples/README.md
@@ -10,16 +10,16 @@ Internal plugin that wraps NeMo Guardrails for PII detection using an Ollama mod
 - See [nemo/README.md](./nemo/README.md) for details
 
 ### nemocheck
-External plugin adapter for NeMo Guardrails check server
-- **Type**: External (separate service)
-- Requires separate NeMo check server deployment
-- See [nemocheck/README.md](./nemocheck/README.md) for details
-
-### nemocheck-internal
 Internal plugin adapter for NeMo Guardrails check server
 - **Type**: Internal
 - Requires separate NeMo check server deployment
-- See [README.md](./nemocheckinternal/README.md) for details
+- See [README.md](./nemocheck/README.md) for details
+
+### nemocheck_external
+External plugin adapter for NeMo Guardrails check server
+- **Type**: External (separate service)
+- Requires separate NeMo check server deployment
+- See [nemocheck_external/README.md](./nemocheck_external/README.md) for details
 
 ## Usage
 
diff --git a/plugins/examples/nemocheck/README.md b/plugins/examples/nemocheck/README.md
index eb064ef..f586c81 100644
--- a/plugins/examples/nemocheck/README.md
+++ b/plugins/examples/nemocheck/README.md
@@ -4,16 +4,15 @@ This directory contains the core `NemoCheck` plugin implementation used by both
 
 ## Prerequisites: Nemo-check server
  * Refer to [orignal repo](https://github.com/m-misiura/demos/tree/main/nemo_openshift/guardrail-checks/deployment) for full instructions
- * Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
+ * Instructions are adapted for the `mcp-gateway` kind cluster to work with an LLM proxy routing to an OpenAI-compatible backend below:
 
-   ```bash
-	docker pull quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1
-	kind load docker-image quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 --name mcp-gateway
-    cd plugins-adapter/plugins/examples/nemocheck/k8deploy
-	kubectl apply -f config-tools.yaml
-	kubectl apply -f server.yaml
+```bash
+docker pull quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1
+kind load docker-image quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 --name mcp-gateway
+cd plugins-adapter/plugins/examples/nemocheck/k8deploy
+make deploy
+```
 
-   ```
 ## Installation
 
 1. Find url of nemo-check-server service. E.g., from svc in `server.yaml`
@@ -23,7 +22,7 @@ This directory contains the core `NemoCheck` plugin implementation used by both
     # plugins/config.yaml - Main plugin configuration file
     plugins:
       - name: "NemoCheck"
-        kind: "plugins.examples.nemocheck.nemocheck.plugin.NemoCheck"
+        kind: "plugins.examples.nemocheck.plugin.NemoCheck"
         description: "Adapter for nemo check server"
         version: "0.1.0"
         hooks: ["tool_pre_invoke", "tool_post_invoke"]
@@ -38,46 +37,56 @@ This directory contains the core `NemoCheck` plugin implementation used by both
 1. In `config.yaml` ensure key `plugins.config.checkserver_url` points to the correct service
 1. Start plugin adapter
 
-## Plugin Development
-
-To install dependencies with dev packages (required for linting and testing):
-
-```bash
-make install-dev
-```
-
-Alternatively, you can also install it in editable mode:
-
-```bash
-make install-editable
-```
-
-## Setting up the development environment
-
-1. Copy .env.template .env
-2. Enable plugins in `.env`
-
 ## Testing
 
 Test modules are created under the `tests` directory.
 
-To run all tests, use the following command:
+To run all tests:
 
 ```bash
-make test
+python -m pytest tests/ -v
 ```
 
 **Note:** To enable logging, set `log_cli = true` in `tests/pytest.ini`.
 
-## Code Linting
-
-Before checking in any code for the project, please lint the code. This can be done using:
-
-```bash
-make lint-fix
+## Test with MCP inspector
+ * Add allowed tools to `plugins-adapter/plugins/examples/nemocheck/k8deploy/config-tools.yaml#check_tool_call_safety`
+<table>
+<tr>
+<th> config-tools.yaml line-127</th>
+<th>Updated to add test2_hello_world </th>
+</tr>
+<tr>
+<td>
+<pre>
+
+```python
+@action(is_system_action=True)
+async def check_tool_call_safety(tool_calls=None, context=None):
+    """Allow list for tool execution."""
+      ...
+      allowed_tools = ["get_weather", "search_web",
+          "get_time", "slack_read_messages"]
+      ...
 ```
+</pre>
+</td>
+<td>
+
+```python
+@action(is_system_action=True)
+async def check_tool_call_safety(tool_calls=None, context=None):
+    """Allow list for tool execution."""
+      ...
+      allowed_tools = ["get_weather", "search_web", "get_time",
+          "test2_hello_world", "slack_read_messages"]
+      ...
+```
+
+</td>
+</tr>
+</table>
 
-# Test
 
-1. Open mcp-inspector to the mcp-gateway
-1. Try running a tool configured/not configured in nemo check config allow list in configmap [E.g.](/plugins/examples/nemocheck/k8deploy/config-tools.yaml)
+ * Redeploy check server
+ * Open the MCP inspector provided by the MCP gateway. Try tools in the allow-list vs. tools not in the allow-list.
diff --git a/plugins/examples/nemocheck/config.yaml b/plugins/examples/nemocheck/config.yaml
index d9f97b1..89b047f 100644
--- a/plugins/examples/nemocheck/config.yaml
+++ b/plugins/examples/nemocheck/config.yaml
@@ -1,6 +1,6 @@
 plugins:
   - name: "NemoCheck"
-    kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2"
+    kind: "plugins.examples.nemocheck.plugin.NemoCheck"
     description: "Nemo Check Adapter"
     version: "0.1.0"
     author: "julianstephen"
diff --git a/plugins/examples/nemocheck_external/k8deploy/Makefile b/plugins/examples/nemocheck/k8deploy/Makefile
similarity index 100%
rename from plugins/examples/nemocheck_external/k8deploy/Makefile
rename to plugins/examples/nemocheck/k8deploy/Makefile
diff --git a/plugins/examples/nemocheck_external/k8deploy/config-tools.yaml b/plugins/examples/nemocheck/k8deploy/config-tools.yaml
similarity index 100%
rename from plugins/examples/nemocheck_external/k8deploy/config-tools.yaml
rename to plugins/examples/nemocheck/k8deploy/config-tools.yaml
diff --git a/plugins/examples/nemocheck_external/k8deploy/server.yaml b/plugins/examples/nemocheck/k8deploy/server.yaml
similarity index 100%
rename from plugins/examples/nemocheck_external/k8deploy/server.yaml
rename to plugins/examples/nemocheck/k8deploy/server.yaml
diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index 598bb4e..46a762c 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -26,17 +26,19 @@
 import logging
 import os
 import requests
+import json
 
 # Initialize logging
 logger = logging.getLogger(__name__)
 log_level = os.getenv("LOGLEVEL", "INFO").upper()
 logger.setLevel(log_level)
 
-MODEL_NAME = os.getenv("NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct")
+MODEL_NAME = os.getenv(
+    "NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct"
+)  # Currently only for logging.
 DEFAULT_CHECK_ENDPOINT = os.getenv(
     "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000"
 )
-
 HEADERS = {
     "Content-Type": "application/json",
 }
@@ -138,8 +140,8 @@ async def tool_pre_invoke(
                 else:
                     metadata = data.get("rails_status")
                     violation = PluginViolation(
-                        reason=f"Tool Check status: {status}",
-                        description="Rails check blocked request",
+                        reason=f"Check tool rails:{status}.",
+                        description=json.dumps(data),
                         code=f"checkserver_http_status_code:{response.status_code}",
                         details=metadata,
                     )
diff --git a/plugins/examples/nemocheck/tests/__init__.py b/plugins/examples/nemocheck/tests/__init__.py
index 53386c4..f4fbc39 100644
--- a/plugins/examples/nemocheck/tests/__init__.py
+++ b/plugins/examples/nemocheck/tests/__init__.py
@@ -1 +1 @@
-"""Tests for nemocheckinternal plugin."""
+"""Tests for nemocheck plugin."""
diff --git a/plugins/examples/nemocheck_external/README.md b/plugins/examples/nemocheck_external/README.md
index 53032ee..94cf97f 100644
--- a/plugins/examples/nemocheck_external/README.md
+++ b/plugins/examples/nemocheck_external/README.md
@@ -6,13 +6,11 @@ This is an external plugin deployment for the NemoCheck guardrails adapter. It r
 
 - **Core Implementation**: `plugins/examples/nemocheck/plugin.py` contains the actual NemoCheck logic
 - **External Deployment**: This directory (`nemocheck_external`) configures and deploys the plugin as an external MCP server
-- **Shared Logic**: Both internal and external deployments use the same underlying implementation to eliminate code duplication
-
 
 ## Run plugin in kind cluster
 
- 1. Run Nemo Guardrails check server. Instructions [here](#Deploy-checkserver)
- 1. Update `CHECK_ENDPOINT` variable in k8deploy/deploy.yaml to point to guardrails check server endpoint
+ 1. Run Nemo Guardrails check server. Instructions are the same as in the internal plugin [here](../nemocheck/README.md#prerequisites-nemo-check-server)
+ 1. Update `CHECK_ENDPOINT` variable in [k8deploy/deploy.yaml](./k8deploy/deploy.yaml) to point to guardrails check server endpoint
 
     ```bash
     cd plugins-adapter/plugins/examples/nemocheck
@@ -39,66 +37,25 @@ This is an external plugin deployment for the NemoCheck guardrails adapter. It r
    make all
    ```
 
-## Test with MCP inspector
- * Add allowed tools to `plugins-adapter/plugins/examples/nemocheck/k8deploy/config-tools.yaml#check_tool_call_safety`
-<table>
-<tr>
-<th> config-tools.yaml line-127</th>
-<th>Updated to add test2_hello_world </th>
-</tr>
-<tr>
-<td>
-<pre>
-
-```python
-@action(is_system_action=True)
-async def check_tool_call_safety(tool_calls=None, context=None):
-    """Allow list for tool execution."""
-      ...
-      allowed_tools = ["get_weather", "search_web",
-          "get_time", "slack_read_messages"]
-      ...
-```
-</pre>
-</td>
-<td>
-
-```python
-@action(is_system_action=True)
-async def check_tool_call_safety(tool_calls=None, context=None):
-    """Allow list for tool execution."""
-      ...
-      allowed_tools = ["get_weather", "search_web", "get_time",
-          "test2_hello_world", "slack_read_messages"]
-      ...
-```
-
-</td>
-</tr>
-</table>
 
+## Plugin Development
 
- * Redeploy check server
- * Open mcp inspector. Try tools in allow list vs tools not in allow list
+To install dependencies with dev packages (required for linting and testing):
 
+```bash
+make install-dev
+```
 
-## Deploy-checkserver
- * Refer to [orignal repo](https://github.com/m-misiura/demos/tree/main/nemo_openshift/guardrail-checks/deployment) for full instructions
- * Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
- * Makefile has targets to load checkserver to kind cluster, etc.
-
-   ```bash
-   cd plugins-adapter/plugins/examples/nemocheck/k8deploy
-   make deploy
-
-   ```
-
+Alternatively, you can also install it in editable mode:
 
-## Testing
+```bash
+make install-editable
+```
 
-Tests are located in the `nemocheck` plugin directory since both deployments share the same core implementation.
+## Setting up the development environment
 
-See the [nemocheck README](../nemocheck/README.md#testing) for testing instructions.
+1. Copy .env.template .env
+2. Enable plugins in `.env`
 
 ## Runtime (server)
 
diff --git a/resources/config/config.yaml b/resources/config/config.yaml
index 4b1530f..13a354b 100644
--- a/resources/config/config.yaml
+++ b/resources/config/config.yaml
@@ -37,20 +37,21 @@ plugins:
       foo: bar
 
   # Nemo Check Example
-  - name: "NemoCheckv2"
-    kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2"
+  - name: "NemoCheck"
+    kind: "plugins.examples.nemocheck.plugin.NemoCheck"
     description: "Adapter for nemo check server"
     version: "0.1.0"
     author: "Julian Stephen"
     config:
-      checkserver_url: "http://nemo-guardrails-service:8000/v1/guardrail/checks"    
+      checkserver_url: "http://nemo-guardrails-service:8000/v1/guardrail/checks"
+
 # Plugin directories to scan
 plugin_dirs:
-  - "plugins/native"           # Built-in plugins
-  - "plugins/custom"           # Custom organization plugins
-  - "/etc/mcpgateway/plugins"  # System-wide plugins
-  - "plugins/examples/nemo"    # Example Nemo guardrails plugins
-  - "plugins/examples/nemocheckinternal"    # Nemo Check Server plugins
+  - "plugins/native"                # Built-in plugins
+  - "plugins/custom"                # Custom organization plugins
+  - "/etc/mcpgateway/plugins"       # System-wide plugins
+  - "plugins/examples/nemo"         # Example Nemo guardrails plugins
+  - "plugins/examples/nemocheck"    # Nemo Check Server plugins
 
 # Global plugin settings
 plugin_settings:
diff --git a/resources/config/nemocheck-internal-config.yaml b/resources/config/nemocheck-internal-config.yaml
index 8650cb1..ccf5238 100644
--- a/resources/config/nemocheck-internal-config.yaml
+++ b/resources/config/nemocheck-internal-config.yaml
@@ -1,8 +1,8 @@
 # plugins/config.yaml - Main plugin configuration file
 plugins:
   # Nemo Check Example
-  - name: "NemoCheckv2"
-    kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2"
+  - name: "NemoCheck"
+    kind: "plugins.examples.nemocheck.plugin.NemoCheck"
     description: "Adapter for nemo check server"
     version: "0.1.0"
     author: "Julian Stephen"
@@ -15,11 +15,11 @@ plugins:
 
 # Plugin directories to scan
 plugin_dirs:
-  - "plugins/native"           # Built-in plugins
-  - "plugins/custom"           # Custom organization plugins
-  - "/etc/mcpgateway/plugins"  # System-wide plugins
-  - "plugins/examples/nemo"    # Example Nemo guardrails plugins
-  - "plugins/examples/nemocheckinternal"    # Nemo Check Server plugins
+  - "plugins/native"                # Built-in plugins
+  - "plugins/custom"                # Custom organization plugins
+  - "/etc/mcpgateway/plugins"       # System-wide plugins
+  - "plugins/examples/nemo"         # Example Nemo guardrails plugins
+  - "plugins/examples/nemocheck"    # Nemo Check Server plugins
 
 # Global plugin settings
 plugin_settings:

From 700a6966aa68798cad2a3b23789762ea0d8c2c85 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 13:15:44 -0700
Subject: [PATCH 13/27] :art: Lint test

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/tests/test_all.py | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/plugins/examples/nemocheck/tests/test_all.py b/plugins/examples/nemocheck/tests/test_all.py
index 85d1c22..3212f77 100644
--- a/plugins/examples/nemocheck/tests/test_all.py
+++ b/plugins/examples/nemocheck/tests/test_all.py
@@ -33,9 +33,13 @@ def plugin_manager():
 async def test_prompt_pre_hook(plugin_manager: PluginManager):
     """Test prompt pre hook across all registered plugins."""
     # Customize payload for testing
-    payload = PromptPrehookPayload(prompt_id="test_prompt", args={"arg0": "This is an argument"})
+    payload = PromptPrehookPayload(
+        prompt_id="test_prompt", args={"arg0": "This is an argument"}
+    )
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(PromptHookType.PROMPT_PRE_FETCH, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        PromptHookType.PROMPT_PRE_FETCH, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing
 
@@ -48,7 +52,9 @@ async def test_prompt_post_hook(plugin_manager: PluginManager):
     prompt_result = PromptResult(messages=[message])
     payload = PromptPosthookPayload(prompt_id="test_prompt", result=prompt_result)
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(PromptHookType.PROMPT_POST_FETCH, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        PromptHookType.PROMPT_POST_FETCH, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing
 
@@ -57,8 +63,12 @@ async def test_prompt_post_hook(plugin_manager: PluginManager):
 async def test_tool_post_hook(plugin_manager: PluginManager):
     """Test tool post hook across all registered plugins."""
     # Customize payload for testing
-    payload = ToolPostInvokePayload(name="test_tool", result={"output0": "output value"})
+    payload = ToolPostInvokePayload(
+        name="test_tool", result={"output0": "output value"}
+    )
     global_context = GlobalContext(request_id="1")
-    result, _ = await plugin_manager.invoke_hook(ToolHookType.TOOL_POST_INVOKE, payload, global_context)
+    result, _ = await plugin_manager.invoke_hook(
+        ToolHookType.TOOL_POST_INVOKE, payload, global_context
+    )
     # Assert expected behaviors
     assert result.continue_processing

From e0b45d773dd5e1822dd33fac9b5aecebc9de0a9b Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 13:20:16 -0700
Subject: [PATCH 14/27] :construction_worker: Update CI references

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .github/workflows/ci.yaml | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 7536ae2..7c3c726 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -42,13 +42,11 @@ jobs:
             git diff --stat
             exit 1
           }
-      - name: Install nemocheck external plugin dependencies
+      - name: Install nemocheck plugin dependencies
         working-directory: ./plugins/examples/nemocheck
         run: |
-          echo "Running nemocheck ext plugin tests.."
+          echo "Running nemocheck plugin tests.."
           uv sync --all-groups
-      - name: Run nemocheck external plugin tests
+      - name: Run nemocheck plugin tests
         working-directory: ./plugins/examples/nemocheck
-        run: uv run pytest tests 
-          
-      # - name: Test
+        run: uv run pytest tests

From b229dbf48bb9b88d13389624d9c1d305146a0bbd Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 13:28:06 -0700
Subject: [PATCH 15/27] :truck::white_check_mark: Update moved nemo check tests

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/{nemocheck_external => nemocheck}/MANIFEST.in  | 0
 .../examples/{nemocheck_external => nemocheck}/pyproject.toml   | 0
 plugins/examples/nemocheck/tests/pytest.ini                     | 2 +-
 plugins/examples/{nemocheck_external => nemocheck}/uv.lock      | 0
 4 files changed, 1 insertion(+), 1 deletion(-)
 rename plugins/examples/{nemocheck_external => nemocheck}/MANIFEST.in (100%)
 rename plugins/examples/{nemocheck_external => nemocheck}/pyproject.toml (100%)
 rename plugins/examples/{nemocheck_external => nemocheck}/uv.lock (100%)

diff --git a/plugins/examples/nemocheck_external/MANIFEST.in b/plugins/examples/nemocheck/MANIFEST.in
similarity index 100%
rename from plugins/examples/nemocheck_external/MANIFEST.in
rename to plugins/examples/nemocheck/MANIFEST.in
diff --git a/plugins/examples/nemocheck_external/pyproject.toml b/plugins/examples/nemocheck/pyproject.toml
similarity index 100%
rename from plugins/examples/nemocheck_external/pyproject.toml
rename to plugins/examples/nemocheck/pyproject.toml
diff --git a/plugins/examples/nemocheck/tests/pytest.ini b/plugins/examples/nemocheck/tests/pytest.ini
index ff60648..e11e911 100644
--- a/plugins/examples/nemocheck/tests/pytest.ini
+++ b/plugins/examples/nemocheck/tests/pytest.ini
@@ -8,6 +8,6 @@ log_format = %(asctime)s [%(module)s] [%(levelname)s] %(message)s
 log_date_format = %Y-%m-%d %H:%M:%S
 addopts = --cov --cov-report term-missing
 env_files = .env
-pythonpath = . src
+pythonpath = .. . src
 filterwarnings =
     ignore::DeprecationWarning:pydantic.*
diff --git a/plugins/examples/nemocheck_external/uv.lock b/plugins/examples/nemocheck/uv.lock
similarity index 100%
rename from plugins/examples/nemocheck_external/uv.lock
rename to plugins/examples/nemocheck/uv.lock

From 33b35654fa4787d318e0f6a176bd6892e2849fcd Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 13:57:32 -0700
Subject: [PATCH 16/27] :wrench: Add hooks to config

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 resources/config/config.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/resources/config/config.yaml b/resources/config/config.yaml
index 13a354b..2b4b048 100644
--- a/resources/config/config.yaml
+++ b/resources/config/config.yaml
@@ -40,6 +40,7 @@ plugins:
   - name: "NemoCheck"
     kind: "plugins.examples.nemocheck.plugin.NemoCheck"
     description: "Adapter for nemo check server"
+    hooks: ["tool_pre_invoke", "tool_post_invoke"]
     version: "0.1.0"
     author: "Julian Stephen"
     config:

From 6ede34f46caceac1fca5e0712d01ca8ba3164eb5 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 15:31:51 -0700
Subject: [PATCH 17/27] :wrench: Updates for deploying external plugin

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .../examples/nemocheck_external/Containerfile |  13 ++-
 plugins/examples/nemocheck_external/Makefile  |   7 +-
 plugins/examples/nemocheck_external/README.md |   7 +-
 .../nemocheck_external/k8deploy/deploy.yaml   |  12 +-
 .../nemocheck_external/pyproject.toml         | 106 ++++++++++++++++++
 .../resources/plugins/config.yaml             |   4 +-
 6 files changed, 134 insertions(+), 15 deletions(-)
 create mode 100644 plugins/examples/nemocheck_external/pyproject.toml

diff --git a/plugins/examples/nemocheck_external/Containerfile b/plugins/examples/nemocheck_external/Containerfile
index 8652213..a834181 100644
--- a/plugins/examples/nemocheck_external/Containerfile
+++ b/plugins/examples/nemocheck_external/Containerfile
@@ -9,6 +9,7 @@ ARG VERSION
 
 ENV APP_HOME=/app
 ENV PLUGINS_TRANSPORT=http
+ENV PYTHONPATH="${HOME}:${PYTHONPATH}"
 
 USER 0
 
@@ -27,8 +28,16 @@ RUN mkdir -p ${APP_HOME} && \
 USER 1001
 
 # Install plugin package
-COPY . .
-RUN pip install --no-cache-dir uv && python -m uv pip install .
+# Note: Build context should be set to parent directory to access ../nemocheck
+# Example: docker build -f nemocheck_external/Containerfile -t nemocheck-external .
+COPY --chown=1001:0 nemocheck ${HOME}/nemocheck
+COPY --chown=1001:0 nemocheck_external ${HOME}/
+
+WORKDIR ${HOME}
+# Install nemocheck package first (the core plugin)
+RUN pip install --no-cache-dir uv && \
+    cd nemocheck && python -m uv pip install . && cd .. && \
+    python -m uv pip install .
 
 # Make default cache directory writable
 RUN mkdir -p -m 0776 ${HOME}/.cache
diff --git a/plugins/examples/nemocheck_external/Makefile b/plugins/examples/nemocheck_external/Makefile
index 6b33011..9f8a139 100644
--- a/plugins/examples/nemocheck_external/Makefile
+++ b/plugins/examples/nemocheck_external/Makefile
@@ -115,9 +115,10 @@ PLATFORM ?= linux/$(shell uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')
 
 container-build:
 	@echo "🔨 Building with $(CONTAINER_RUNTIME) for platform $(PLATFORM)..."
-	$(CONTAINER_RUNTIME) build \
+	@echo "📁 Using parent directory as build context to include nemocheck plugin..."
+	cd .. && $(CONTAINER_RUNTIME) build \
 		--platform=$(PLATFORM) \
-		-f $(CONTAINER_FILE) \
+		-f nemocheck_external/$(CONTAINER_FILE) \
 		--tag $(IMAGE_BASE):$(IMAGE_TAG) \
 		.
 	@echo "✅ Built image: $(call get_image_name)"
@@ -418,7 +419,7 @@ load: container-build
 	kind load docker-image  $(IMAGE_BASE):$(IMAGE_TAG) --name mcp-gateway
 
 .PHONY: redeploy
-redeploy: 
+redeploy:
 	-kubectl delete -f k8deploy/deploy.yaml
 	kubectl apply -f k8deploy/deploy.yaml
 
diff --git a/plugins/examples/nemocheck_external/README.md b/plugins/examples/nemocheck_external/README.md
index 94cf97f..e32c59b 100644
--- a/plugins/examples/nemocheck_external/README.md
+++ b/plugins/examples/nemocheck_external/README.md
@@ -6,6 +6,8 @@ This is an external plugin deployment for the NemoCheck guardrails adapter. It r
 
 - **Core Implementation**: `plugins/examples/nemocheck/plugin.py` contains the actual NemoCheck logic
 - **External Deployment**: This directory (`nemocheck_external`) configures and deploys the plugin as an external MCP server
+- **Shared Logic**: The config references `nemocheck` to use the core implementation without code duplication
+- **Container Build**: The Containerfile copies the `nemocheck` directory during build to include the core implementation
 
 ## Run plugin in kind cluster
 
@@ -13,7 +15,7 @@ This is an external plugin deployment for the NemoCheck guardrails adapter. It r
  1. Update `CHECK_ENDPOINT` variable in [k8deploy/deploy.yaml](./k8deploy/deploy.yaml) to point to guardrails check server endpoint
 
     ```bash
-    cd plugins-adapter/plugins/examples/nemocheck
+    cd plugins-adapter/plugins/examples/nemocheck_external
     make deploy
     ```
  1.
@@ -21,7 +23,8 @@ This is an external plugin deployment for the NemoCheck guardrails adapter. It r
     <summary>Non-kind k8 cluster instructions</summary>
 
     ```bash
-        cd plugins-adapter/plugins/examples/nemocheck
+        cd plugins-adapter/plugins/examples/nemocheck_external
+        make deploy
         make container-build
         # push image to your container repo and update image name in k8deploy/deploy.yaml
         kubectl apply -f k8deploy/deploy.yaml
diff --git a/plugins/examples/nemocheck_external/k8deploy/deploy.yaml b/plugins/examples/nemocheck_external/k8deploy/deploy.yaml
index 5354549..4ae0703 100644
--- a/plugins/examples/nemocheck_external/k8deploy/deploy.yaml
+++ b/plugins/examples/nemocheck_external/k8deploy/deploy.yaml
@@ -39,17 +39,17 @@ spec:
               value: "0.0.0.0"
             - name: PLUGINS_ENABLED
               value: "false"
-            - name: PLUGINS_CLI_COMPLETION  
+            - name: PLUGINS_CLI_COMPLETION
               value: "false"
-            - name: PLUGINS_CLI_MARKUP_MODE  
+            - name: PLUGINS_CLI_MARKUP_MODE
               value: "rich"
-            - name: PLUGINS_CONFIG  
+            - name: PLUGINS_CONFIG
               value: "./resources/plugins/config.yaml"
-            # - name: CHUK_MCP_CONFIG_PATH  
+            # - name: CHUK_MCP_CONFIG_PATH
               # value: "./resources/runtime/config.yaml"
-            - name: MCP_SSL_ENABLED  
+            - name: MCP_SSL_ENABLED
               value: "false"
-            - name: MCP_SSL_CERT_REQS  
+            - name: MCP_SSL_CERT_REQS
               value: "0"
             - name: LOGLEVEL
               value: "DEBUG"
diff --git a/plugins/examples/nemocheck_external/pyproject.toml b/plugins/examples/nemocheck_external/pyproject.toml
new file mode 100644
index 0000000..142de73
--- /dev/null
+++ b/plugins/examples/nemocheck_external/pyproject.toml
@@ -0,0 +1,106 @@
+# ----------------------------------------------------------------
+# 💡 Build system (PEP 517)
+#     - setuptools ≥ 77 gives SPDX licence support (PEP 639)
+#     - wheel is needed by most build front-ends
+# ----------------------------------------------------------------
+[build-system]
+requires = ["setuptools>=77", "wheel"]
+build-backend = "setuptools.build_meta"
+
+# ----------------------------------------------------------------
+# 📦 Core project metadata (PEP 621)
+# ----------------------------------------------------------------
+[project]
+name = "nemocheck_external"
+version = "0.1.0"
+description = "Nemo-Check guardrails external"
+keywords = ["MCP","API","gateway","tools",
+  "agents","agentic ai","model context protocol","multi-agent","fastapi",
+  "json-rpc","sse","websocket","federation","security","authentication"
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Framework :: FastAPI",
+    "Framework :: AsyncIO",
+    "Topic :: Internet :: WWW/HTTP :: WSGI :: Application",
+    "Topic :: Software Development :: Libraries :: Application Frameworks"
+]
+readme = "README.md"
+requires-python = ">=3.11,<3.14"
+license = "Apache-2.0"
+license-files = ["LICENSE"]
+
+maintainers = [
+    {name = "julianstephen", email = "julian.stephen@gmail.com"}
+]
+
+authors = [
+    {name = "julianstephen", email = "julian.stephen@gmail.com"}
+]
+
+dependencies = [
+    "mcp>=1.16.0",
+    "mcp-contextforge-gateway",
+]
+
+# URLs
+[project.urls]
+Homepage = "https://ibm.github.io/mcp-context-forge/"
+Documentation = "https://ibm.github.io/mcp-context-forge/"
+Repository = "https://github.com/IBM/mcp-context-forge"
+"Bug Tracker" = "https://github.com/IBM/mcp-context-forge/issues"
+Changelog = "https://github.com/IBM/mcp-context-forge/blob/main/CHANGELOG.md"
+
+[tool.uv.sources]
+mcp-contextforge-gateway = { git = "https://github.com/IBM/mcp-context-forge.git", rev = "main" }
+
+# ----------------------------------------------------------------
+# Optional dependency groups (extras)
+# ----------------------------------------------------------------
+[project.optional-dependencies]
+dev = [
+    "black>=25.1.0",
+    "pytest>=8.4.1",
+    "pytest-asyncio>=1.1.0",
+    "pytest-cov>=6.2.1",
+    "pytest-dotenv>=0.5.2",
+    "pytest-env>=1.1.5",
+    "pytest-examples>=0.0.18",
+    "pytest-md-report>=0.7.0",
+    "pytest-rerunfailures>=15.1",
+    "pytest-trio>=0.8.0",
+    "pytest-xdist>=3.8.0",
+    "ruff>=0.12.9",
+    "unimport>=1.2.1",
+    "uv>=0.8.11",
+]
+
+# --------------------------------------------------------------------
+#  🔧 setuptools-specific configuration
+# --------------------------------------------------------------------
+[tool.setuptools]
+include-package-data = true          # ensure wheels include the data files
+
+# Automatic discovery: keep every package that starts with "nemocheck_external"
+[tool.setuptools.packages.find]
+include = ["nemocheck_external*"]
+exclude = ["tests*"]
+
+##  Runtime data files  ------------------------------------------------
+[tool.setuptools.package-data]
+nemocheck_external = [
+  "resources/plugins/config.yaml",
+]
+
+[dependency-groups]
+dev = [
+    "pytest>=8.4.2",
+    "pytest-asyncio>=1.3.0",
+    "pytest-cov>=7.0.0",
+    "ruff>=0.14.14",
+]
diff --git a/plugins/examples/nemocheck_external/resources/plugins/config.yaml b/plugins/examples/nemocheck_external/resources/plugins/config.yaml
index bf3f0db..d4abf3b 100644
--- a/plugins/examples/nemocheck_external/resources/plugins/config.yaml
+++ b/plugins/examples/nemocheck_external/resources/plugins/config.yaml
@@ -1,6 +1,6 @@
 plugins:
   - name: "NemoCheck"
-    kind: "plugin.NemoCheck"
+    kind: "nemocheck.plugin.NemoCheck"
     description: "Adapter for Nemo-Check guardrails"
     version: "0.1.0"
     author: "julianstephen"
@@ -17,7 +17,7 @@ plugins:
 
 # Plugin directories to scan
 plugin_dirs:
-  - "../nemocheck"
+  - "."
 
 # Global plugin settings
 plugin_settings:

From 580262f08b6c63bffa300811083c7b00da260723 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 15:33:28 -0700
Subject: [PATCH 18/27] :goal_net: Add error handling for empty config

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/plugin.py | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index 46a762c..1751590 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -37,7 +37,7 @@
     "NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct"
 )  # Currently only for logging.
 DEFAULT_CHECK_ENDPOINT = os.getenv(
-    "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000"
+    "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000/v1/guardrail/checks"
 )
 HEADERS = {
     "Content-Type": "application/json",
@@ -55,9 +55,14 @@ def __init__(self, config: PluginConfig):
         """
         super().__init__(config)
         # Allow config to override the endpoint
-        self.check_endpoint = config.config.get(
-            "checkserver_url", DEFAULT_CHECK_ENDPOINT
-        )
+        # Handle case where config.config might be None or empty
+        if config.config and isinstance(config.config, dict):
+            self.check_endpoint = config.config.get(
+                "checkserver_url", DEFAULT_CHECK_ENDPOINT
+            )
+        else:
+            self.check_endpoint = DEFAULT_CHECK_ENDPOINT
+            logger.warning("Plugin config is empty or invalid, using default endpoint")
         logger.info(f"Nemo Check endpoint: {self.check_endpoint}")
 
     async def prompt_pre_fetch(

From aee2006772d37849a554ec8ef0f535b24515c994 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 15:41:19 -0700
Subject: [PATCH 19/27] :fire: Minimal pyproject

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .../nemocheck_external/pyproject.toml         | 81 +------------------
 1 file changed, 4 insertions(+), 77 deletions(-)

diff --git a/plugins/examples/nemocheck_external/pyproject.toml b/plugins/examples/nemocheck_external/pyproject.toml
index 142de73..4827824 100644
--- a/plugins/examples/nemocheck_external/pyproject.toml
+++ b/plugins/examples/nemocheck_external/pyproject.toml
@@ -13,94 +13,21 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "nemocheck_external"
 version = "0.1.0"
-description = "Nemo-Check guardrails external"
-keywords = ["MCP","API","gateway","tools",
-  "agents","agentic ai","model context protocol","multi-agent","fastapi",
-  "json-rpc","sse","websocket","federation","security","authentication"
-]
-classifiers = [
-    "Development Status :: 4 - Beta",
-    "Intended Audience :: Developers",
-    "Programming Language :: Python :: 3",
-    "Programming Language :: Python :: 3.11",
-    "Programming Language :: Python :: 3.12",
-    "Programming Language :: Python :: 3.13",
-    "Framework :: FastAPI",
-    "Framework :: AsyncIO",
-    "Topic :: Internet :: WWW/HTTP :: WSGI :: Application",
-    "Topic :: Software Development :: Libraries :: Application Frameworks"
-]
-readme = "README.md"
+description = "NemoCheck external plugin"
 requires-python = ">=3.11,<3.14"
-license = "Apache-2.0"
-license-files = ["LICENSE"]
-
-maintainers = [
-    {name = "julianstephen", email = "julian.stephen@gmail.com"}
-]
-
-authors = [
-    {name = "julianstephen", email = "julian.stephen@gmail.com"}
-]
 
 dependencies = [
     "mcp>=1.16.0",
     "mcp-contextforge-gateway",
 ]
 
-# URLs
-[project.urls]
-Homepage = "https://ibm.github.io/mcp-context-forge/"
-Documentation = "https://ibm.github.io/mcp-context-forge/"
-Repository = "https://github.com/IBM/mcp-context-forge"
-"Bug Tracker" = "https://github.com/IBM/mcp-context-forge/issues"
-Changelog = "https://github.com/IBM/mcp-context-forge/blob/main/CHANGELOG.md"
-
 [tool.uv.sources]
 mcp-contextforge-gateway = { git = "https://github.com/IBM/mcp-context-forge.git", rev = "main" }
 
-# ----------------------------------------------------------------
-# Optional dependency groups (extras)
-# ----------------------------------------------------------------
-[project.optional-dependencies]
-dev = [
-    "black>=25.1.0",
-    "pytest>=8.4.1",
-    "pytest-asyncio>=1.1.0",
-    "pytest-cov>=6.2.1",
-    "pytest-dotenv>=0.5.2",
-    "pytest-env>=1.1.5",
-    "pytest-examples>=0.0.18",
-    "pytest-md-report>=0.7.0",
-    "pytest-rerunfailures>=15.1",
-    "pytest-trio>=0.8.0",
-    "pytest-xdist>=3.8.0",
-    "ruff>=0.12.9",
-    "unimport>=1.2.1",
-    "uv>=0.8.11",
-]
-
 # --------------------------------------------------------------------
 #  🔧 setuptools-specific configuration
 # --------------------------------------------------------------------
 [tool.setuptools]
-include-package-data = true          # ensure wheels include the data files
-
-# Automatic discovery: keep every package that starts with "nemocheck_external"
-[tool.setuptools.packages.find]
-include = ["nemocheck_external*"]
-exclude = ["tests*"]
-
-##  Runtime data files  ------------------------------------------------
-[tool.setuptools.package-data]
-nemocheck_external = [
-  "resources/plugins/config.yaml",
-]
-
-[dependency-groups]
-dev = [
-    "pytest>=8.4.2",
-    "pytest-asyncio>=1.3.0",
-    "pytest-cov>=7.0.0",
-    "ruff>=0.14.14",
-]
+# No packages - this is just a runtime environment
+# Plugin code comes from the nemocheck package
+packages = []

From c6d83f2f81acebc8bf68f19697a3a1a1de68e6e9 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 17:11:43 -0700
Subject: [PATCH 20/27] :art: Lint plugin

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/plugin.py | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index 1751590..f7b58fd 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -62,7 +62,9 @@ def __init__(self, config: PluginConfig):
             )
         else:
             self.check_endpoint = DEFAULT_CHECK_ENDPOINT
-            logger.warning("Plugin config is empty or invalid, using default endpoint")
+            logger.warning(
+                "Plugin config is empty or invalid, using default endpoint"
+            )
         logger.info(f"Nemo Check endpoint: {self.check_endpoint}")
 
     async def prompt_pre_fetch(
@@ -119,7 +121,9 @@ async def tool_pre_invoke(
                             "type": "function",
                             "function": {
                                 "name": tool_name,
-                                "arguments": payload.args.get("tool_args", None),
+                                "arguments": payload.args.get(
+                                    "tool_args", None
+                                ),
                             },
                         }
                     ],
@@ -174,7 +178,9 @@ async def tool_pre_invoke(
                 code="checkserver_connection_error",
                 details={},
             )
-            return ToolPreInvokeResult(continue_processing=False, violation=violation)
+            return ToolPreInvokeResult(
+                continue_processing=False, violation=violation
+            )
 
     async def tool_post_invoke(
         self, payload: ToolPostInvokePayload, context: PluginContext

From b00c91cfa0f541897f98a5fddf6f055a6f1656c4 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 17:24:27 -0700
Subject: [PATCH 21/27] :white_check_mark::fire: Remove old plugin with
 refactor and update tests

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 .../examples/nemocheck/nemocheck/plugin.py    | 256 ------------------
 plugins/examples/nemocheck/plugin.py          |  82 +++++-
 .../nemocheck/tests/test_nemocheck.py         |  18 +-
 3 files changed, 89 insertions(+), 267 deletions(-)
 delete mode 100644 plugins/examples/nemocheck/nemocheck/plugin.py

diff --git a/plugins/examples/nemocheck/nemocheck/plugin.py b/plugins/examples/nemocheck/nemocheck/plugin.py
deleted file mode 100644
index 4983795..0000000
--- a/plugins/examples/nemocheck/nemocheck/plugin.py
+++ /dev/null
@@ -1,256 +0,0 @@
-"""Adapter for Nemo-Check guardrails.
-
-Copyright 2025
-SPDX-License-Identifier: Apache-2.0
-Authors: julianstephen
-
-This module loads configurations for plugins.
-"""
-
-# First-Party
-from mcpgateway.plugins.framework import (
-    Plugin,
-    PluginConfig,
-    PluginContext,
-    PromptPosthookPayload,
-    PromptPosthookResult,
-    PromptPrehookPayload,
-    PromptPrehookResult,
-    ToolPostInvokePayload,
-    ToolPostInvokeResult,
-    ToolPreInvokePayload,
-    ToolPreInvokeResult,
-    PluginViolation,
-)
-
-
-import logging
-import os
-import requests
-
-headers = {
-    "Content-Type": "application/json",
-}
-# Initialize logging service first
-logger = logging.getLogger(__name__)
-log_level = os.getenv("LOGLEVEL", "INFO").upper()
-logger.setLevel(log_level)
-
-MODEL_NAME = os.getenv(
-    "NEMO_MODEL", "meta-llama/llama-3-3-70b-instruct"
-)  # Currently only for logging.
-CHECK_ENDPOINT = os.getenv(
-    "CHECK_ENDPOINT", "http://nemo-guardrails-service:8000"
-)
-
-
-class NemoCheck(Plugin):
-    """Adapter for Nemo-Check guardrails."""
-
-    def __init__(self, config: PluginConfig):
-        """Entry init block for plugin.
-
-        Args:
-          logger: logger that the skill can make use of
-          config: the skill configuration
-        """
-        super().__init__(config)
-
-    async def prompt_pre_fetch(
-        self, payload: PromptPrehookPayload, context: PluginContext
-    ) -> PromptPrehookResult:
-        """The plugin hook run before a prompt is retrieved and rendered.
-
-        Args:
-            payload: The prompt payload to be analyzed.
-            context: contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the prompt can proceed.
-        """
-
-        return PromptPrehookResult(continue_processing=True)
-
-    async def prompt_post_fetch(
-        self, payload: PromptPosthookPayload, context: PluginContext
-    ) -> PromptPosthookResult:
-        """Plugin hook run after a prompt is rendered.
-
-        Args:
-            payload: The prompt payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the prompt can proceed.
-        """
-        return PromptPosthookResult(continue_processing=True)
-
-    async def tool_pre_invoke(
-        self, payload: ToolPreInvokePayload, context: PluginContext
-    ) -> ToolPreInvokeResult:
-        """Plugin hook run before a tool is invoked.
-
-        Args:
-            payload: The tool payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the tool can proceed.
-        """
-        logger.info("[NemoCheck] Starting tool_pre_invoke")
-        logger.info(payload)
-        tool_name = payload.name  # ("tool_name", None)
-        check_nemo_payload = {
-            "model": MODEL_NAME,  # ideally optional
-            "messages": [
-                {
-                    "role": "assistant",
-                    "tool_calls": [
-                        {
-                            "id": "call_plug_adap_nem_check_123",
-                            "type": "function",
-                            "function": {
-                                "name": tool_name,
-                                "arguments": payload.args.get(
-                                    "tool_args", None
-                                ),
-                            },
-                        }
-                    ],
-                }
-            ],
-        }
-        violation = None
-        response = requests.post(
-            CHECK_ENDPOINT, headers=headers, json=check_nemo_payload
-        )
-        if response.status_code == 200:
-            data = response.json()
-            status = data.get("status", "blocked")
-            logger.debug(f"rails reply:{data}")
-            if status == "success":
-                metadata = data.get("rails_status")
-                result = ToolPreInvokeResult(
-                    continue_processing=True, metadata=metadata
-                )
-            else:
-                metadata = data.get("rails_status")
-                violation = PluginViolation(
-                    reason=f"Tool Check status:{status}",
-                    description="Rails check blocked request",
-                    code=f"checkserver_http_status_code:{response.status_code}",
-                    details=metadata,
-                )
-                result = ToolPreInvokeResult(
-                    continue_processing=False,
-                    violation=violation,
-                    metadata=metadata,
-                )
-
-        else:
-            violation = PluginViolation(
-                reason="Tool Check Unavailable",
-                description="Tool arguments check server returned error:",
-                code=f"checkserver_http_status_code:{response.status_code}",
-                details={},
-            )
-            result = ToolPreInvokeResult(
-                continue_processing=False, violation=violation
-            )
-        logger.info(response)
-
-        return result
-
-    async def tool_post_invoke(
-        self, payload: ToolPostInvokePayload, context: PluginContext
-    ) -> ToolPostInvokeResult:
-        """Plugin hook run after a tool is invoked.
-
-        Args:
-            payload: The tool result payload to be analyzed.
-            context: Contextual information about the hook call.
-
-        Returns:
-            The result of the plugin's analysis, including whether the tool result should proceed.
-        """
-        logger.info(
-            f"[NemoCheck] Starting tool post invoke hook with payload {payload}"
-        )
-
-        # Extract content from payload.result
-        # payload.result format: {'content': [{'type': 'text', 'text': 'Hello, bob!'}]}
-        result_content = payload.result.get("content", [])
-        tool_name = payload.name
-
-        if not result_content:
-            logger.warning(
-                "[NemoCheck] No content in tool result, skipping check"
-            )
-            return ToolPostInvokeResult(continue_processing=True)
-
-        # Extract text content from the content array
-        # TODO: what to do if there's actually multiple texts?
-        text_content = ""
-        for item in result_content:
-            if item.get("type") == "text":
-                text_content += item.get("text", "")
-
-        # Build NeMo check payload for tool response
-        check_nemo_payload = {
-            "model": MODEL_NAME,  # ideally optional
-            "messages": [
-                {"role": "tool", "content": text_content, "name": tool_name}
-            ],
-        }
-
-        logger.debug(
-            f"[NemoCheck] Payload for guardrail check: {check_nemo_payload}"
-        )
-
-        violation = None
-        try:
-            response = requests.post(
-                CHECK_ENDPOINT, headers=headers, json=check_nemo_payload
-            )
-            if response.status_code == 200:
-                data = response.json()
-                status = data.get("status", "blocked")
-                logger.debug(f"[NemoCheck] Rails reply: {data}")
-
-                if status == "success":
-                    metadata = data.get("rails_status")
-                    result = ToolPostInvokeResult(
-                        continue_processing=True, metadata=metadata
-                    )
-                else:  # blocked
-                    metadata = data.get("rails_status")
-                    violation = PluginViolation(
-                        reason=f"Tool response check status: {status}",
-                        description="Rails check blocked tool response",
-                        code=f"checkserver_http_status_code:{response.status_code}",
-                        details=metadata,
-                    )
-                    result = ToolPostInvokeResult(
-                        continue_processing=False,
-                        violation=violation,
-                        metadata=metadata,
-                    )
-            else:
-                violation = PluginViolation(
-                    reason="Tool response check unavailable",
-                    description="Tool response check server returned error",
-                    code=f"checkserver_http_status_code:{response.status_code}",
-                    details={},
-                )
-                result = ToolPostInvokeResult(
-                    continue_processing=False, violation=violation
-                )
-
-            logger.info(f"[NemoCheck] Tool post invoke result: {result}")
-            return result
-
-        except Exception as e:
-            logger.error(f"[NemoCheck] Error checking tool response: {e}")
-            return ToolPostInvokeResult(
-                continue_processing=True
-            )  # Fail open on error
diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index f7b58fd..8a6eaee 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -194,4 +194,84 @@ async def tool_post_invoke(
         Returns:
             The result of the plugin's analysis, including whether the tool result should proceed.
         """
-        return ToolPostInvokeResult(continue_processing=True)
+        logger.info(
+            f"[NemoCheck] Starting tool post invoke hook with payload {payload}"
+        )
+
+        # Extract content from payload.result
+        # payload.result format: {'content': [{'type': 'text', 'text': 'Hello, bob!'}]}
+        result_content = payload.result.get("content", [])
+        tool_name = payload.name
+
+        if not result_content:
+            logger.warning(
+                "[NemoCheck] No content in tool result, skipping check"
+            )
+            return ToolPostInvokeResult(continue_processing=True)
+
+        # Extract text content from the content array
+        # TODO: what to do if there's actually multiple texts?
+        text_content = ""
+        for item in result_content:
+            if item.get("type") == "text":
+                text_content += item.get("text", "")
+
+        # Build NeMo check payload for tool response
+        check_nemo_payload = {
+            "model": MODEL_NAME,  # ideally optional
+            "messages": [
+                {"role": "tool", "content": text_content, "name": tool_name}
+            ],
+        }
+
+        logger.debug(
+            f"[NemoCheck] Payload for guardrail check: {check_nemo_payload}"
+        )
+
+        violation = None
+        try:
+            response = requests.post(
+                self.check_endpoint, headers=HEADERS, json=check_nemo_payload
+            )
+            if response.status_code == 200:
+                data = response.json()
+                status = data.get("status", "blocked")
+                logger.debug(f"[NemoCheck] Rails reply: {data}")
+
+                if status == "success":
+                    metadata = data.get("rails_status")
+                    result = ToolPostInvokeResult(
+                        continue_processing=True, metadata=metadata
+                    )
+                else:  # blocked
+                    metadata = data.get("rails_status")
+                    violation = PluginViolation(
+                        reason=f"Tool response check status: {status}",
+                        description="Rails check blocked tool response",
+                        code=f"checkserver_http_status_code:{response.status_code}",
+                        details=metadata,
+                    )
+                    result = ToolPostInvokeResult(
+                        continue_processing=False,
+                        violation=violation,
+                        metadata=metadata,
+                    )
+            else:
+                violation = PluginViolation(
+                    reason="Tool response check unavailable",
+                    description="Tool response check server returned error",
+                    code=f"checkserver_http_status_code:{response.status_code}",
+                    details={},
+                )
+                result = ToolPostInvokeResult(
+                    continue_processing=False, violation=violation
+                )
+
+            logger.info(f"[NemoCheck] Tool post invoke result: {result}")
+            return result
+
+        except Exception as e:
+            logger.error(f"[NemoCheck] Error checking tool response: {e}")
+            return ToolPostInvokeResult(
+                continue_processing=True
+            )  # Fail open on error
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 10a7e7f..9fe24c1 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -9,6 +9,7 @@
 # First-Party
 from mcpgateway.plugins.framework import (
     PluginConfig,
+    PluginContext,
     GlobalContext,
     PromptPrehookPayload,
     ToolPostInvokePayload,
@@ -33,8 +34,8 @@ def plugin():
 
 @pytest.fixture
 def context():
-    """Create a GlobalContext instance."""
-    return GlobalContext(request_id="1")
+    """Create a PluginContext instance."""
+    return PluginContext(global_context=GlobalContext(request_id="1"))
 
 
 def mock_http_response(status_code, response_data=None):
@@ -52,7 +53,6 @@ async def test_prompt_pre_fetch(plugin, context):
     payload = PromptPrehookPayload(
         prompt_id="test_prompt", args={"arg0": "This is an argument"}
     )
-    context = GlobalContext(request_id="1")
     result = await plugin.prompt_pre_fetch(payload, context)
     assert result.continue_processing
 
@@ -99,7 +99,7 @@ async def test_tool_pre_invoke_scenarios(
     )
 
     with patch(
-        "nemocheck.plugin.requests.post",
+        "plugin.requests.post",
         return_value=mock_http_response(status_code, response_data),
     ):
         result = await plugin.tool_pre_invoke(payload, context)
@@ -150,7 +150,7 @@ async def test_tool_post_invoke_http_scenarios(
     )
 
     with patch(
-        "nemocheck.plugin.requests.post",
+        "plugin.requests.post",
         return_value=mock_http_response(status_code, response_data),
     ):
         result = await plugin.tool_post_invoke(payload, context)
@@ -191,7 +191,7 @@ async def test_tool_post_invoke_concatenates_text(plugin, context):
     )
 
     with patch(
-        "nemocheck.plugin.requests.post",
+        "plugin.requests.post",
         return_value=mock_http_response(
             200, {"status": "success", "rails_status": {}}
         ),
@@ -217,7 +217,7 @@ async def test_tool_post_invoke_filters_non_text(plugin, context):
     )
 
     with patch(
-        "nemocheck.plugin.requests.post",
+        "plugin.requests.post",
         return_value=mock_http_response(
             200, {"status": "success", "rails_status": {}}
         ),
@@ -237,9 +237,7 @@ async def test_tool_post_invoke_fails_open_on_exception(plugin, context):
         result={"content": [{"type": "text", "text": "content"}]},
     )
 
-    with patch(
-        "nemocheck.plugin.requests.post", side_effect=Exception("Network error")
-    ):
+    with patch("plugin.requests.post", side_effect=Exception("Network error")):
         result = await plugin.tool_post_invoke(payload, context)
 
     assert result.continue_processing

From 1354e0e6a576d624c283554607306c095eb492b8 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 13 Feb 2026 17:29:45 -0700
Subject: [PATCH 22/27] :loud_sound: Update plugin log

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/plugin.py | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index 8a6eaee..cc87980 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -107,9 +107,11 @@ async def tool_pre_invoke(
         Returns:
             The result of the plugin's analysis, including whether the tool can proceed.
         """
-        logger.info("tool_pre_invoke....")
-        logger.info(payload)
-        tool_name = payload.name  # ("tool_name", None)
+        logger.info(
+            f"[NemoCheck] Starting tool pre invoke hook with payload {payload}"
+        )
+
+        tool_name = payload.name
         check_nemo_payload = {
             "model": MODEL_NAME,
             "messages": [
@@ -139,7 +141,7 @@ async def tool_pre_invoke(
             if response.status_code == 200:
                 data = response.json()
                 status = data.get("status", "blocked")
-                logger.debug(f"rails reply: {data}")
+                logger.debug(f"[NemoCheck] Rails reply: {data}")
 
                 if status == "success":
                     metadata = data.get("rails_status")
@@ -171,7 +173,7 @@ async def tool_pre_invoke(
                 )
 
         except Exception as e:
-            logger.error(f"Error calling Nemo Check endpoint: {e}")
+            logger.error(f"[NemoCheck] Error checking tool arguments: {e}")
             violation = PluginViolation(
                 reason="Tool Check Error",
                 description=f"Failed to connect to check server: {str(e)}",

From 0f710aa1fbf2f790f7bb95b3d85856429f4b94f7 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Mon, 16 Feb 2026 10:02:06 -0700
Subject: [PATCH 23/27] :white_check_mark: Update server tests

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 tests/test_server.py | 119 ++++++++++++++++++++++---------------------
 1 file changed, 61 insertions(+), 58 deletions(-)

diff --git a/tests/test_server.py b/tests/test_server.py
index efc0fe7..64dcfe6 100644
--- a/tests/test_server.py
+++ b/tests/test_server.py
@@ -206,10 +206,25 @@ async def test_getToolPostInvokeResponse_modified_payload(
     # Inject mock manager
     src.server.manager = mock_manager
 
-    # Call the function
-    response = await src.server.getToolPostInvokeResponse(
-        sample_tool_result_body
-    )
+    # Spy on json.dumps to capture what body is being serialized
+    original_dumps = json.dumps
+    captured_body = None
+
+    def spy_dumps(obj, **kwargs):
+        nonlocal captured_body
+        # Capture the body dict that's being serialized
+        if isinstance(obj, dict) and "result" in obj and "jsonrpc" in obj:
+            captured_body = obj
+        return original_dumps(obj, **kwargs)
+
+    json.dumps = spy_dumps
+    try:
+        # Call the function
+        response = await src.server.getToolPostInvokeResponse(
+            sample_tool_result_body
+        )
+    finally:
+        json.dumps = original_dumps
 
     # Verify the hook was called
     assert mock_manager.invoke_hook.called
@@ -217,6 +232,18 @@ async def test_getToolPostInvokeResponse_modified_payload(
     # Verify response was created
     assert response is not None
 
+    # Verify the body was modified with the new result
+    assert captured_body is not None, (
+        "json.dumps should have been called with the modified body"
+    )
+    assert captured_body["result"] == modified_result
+    assert (
+        captured_body["result"]["content"][0]["text"] == "Modified tool result"
+    )
+    # Verify original metadata (jsonrpc, id) is preserved
+    assert captured_body["jsonrpc"] == sample_tool_result_body["jsonrpc"]
+    assert captured_body["id"] == sample_tool_result_body["id"]
+
 
 @pytest.mark.asyncio
 async def test_getToolPostInvokeResponse_multiple_content_items(
@@ -318,49 +345,6 @@ async def test_process_response_body_buffer_with_sse_format(
     assert response is not None
 
 
-@pytest.mark.asyncio
-async def test_process_response_body_buffer_empty(
-    mock_envoy_modules, mock_manager
-):
-    """Test process_response_body_buffer with empty buffer."""
-    setup_response_mocks(mock_envoy_modules)
-    import src.server
-
-    src.server.manager = mock_manager
-    response = await src.server.process_response_body_buffer(bytearray())
-
-    # Verify ProcessingResponse was returned
-    assert response is not None
-    assert not mock_manager.invoke_hook.called, (
-        "Tool post-invoke hook should not be called for empty buffer"
-    )
-
-
-@pytest.mark.asyncio
-async def test_process_response_body_buffer_non_tool_result(
-    mock_envoy_modules, mock_manager
-):
-    """Test process_response_body_buffer with non-tool result (error response)."""
-    setup_response_mocks(mock_envoy_modules)
-    import src.server
-
-    src.server.manager = mock_manager
-
-    error_response = {
-        "jsonrpc": "2.0",
-        "id": "test-error",
-        "error": {"code": -32000, "message": "Error"},
-    }
-    buffer = bytearray(json.dumps(error_response).encode("utf-8"))
-    response = await src.server.process_response_body_buffer(buffer)
-
-    # Verify ProcessingResponse was returned
-    assert response is not None
-    assert not mock_manager.invoke_hook.called, (
-        "Tool post-invoke hook should not be called for error responses"
-    )
-
-
 @pytest.mark.asyncio
 async def test_process_response_body_buffer_multiple_chunks_scenario(
     mock_envoy_modules, mock_manager
@@ -398,26 +382,45 @@ async def test_process_response_body_buffer_multiple_chunks_scenario(
 
 
 @pytest.mark.asyncio
-async def test_process_response_body_buffer_single_chunk_with_end_of_stream(
+async def test_process_response_body_buffer_empty(
     mock_envoy_modules, mock_manager
 ):
-    """Test buffering: all content in one chunk with end_of_stream."""
+    """Test process_response_body_buffer with empty buffer."""
     setup_response_mocks(mock_envoy_modules)
     import src.server
 
-    setup_manager_with_result(mock_manager)
     src.server.manager = mock_manager
+    response = await src.server.process_response_body_buffer(bytearray())
 
-    tool_result = {
+    # Verify hook is NOT called for empty buffer
+    assert not mock_manager.invoke_hook.called, (
+        "Tool post-invoke hook should not be called for empty buffer"
+    )
+    # Verify response is returned (function doesn't crash on empty buffer)
+    assert response is not None
+
+
+@pytest.mark.asyncio
+async def test_process_response_body_buffer_non_tool_result(
+    mock_envoy_modules, mock_manager
+):
+    """Test process_response_body_buffer with non-tool result (error response)."""
+    setup_response_mocks(mock_envoy_modules)
+    import src.server
+
+    src.server.manager = mock_manager
+
+    error_response = {
         "jsonrpc": "2.0",
-        "id": "test-single",
-        "result": {"content": [{"type": "text", "text": "Single chunk"}]},
+        "id": "test-error",
+        "error": {"code": -32000, "message": "Error"},
     }
-    buffer = bytearray(json.dumps(tool_result).encode("utf-8"))
+    buffer = bytearray(json.dumps(error_response).encode("utf-8"))
     response = await src.server.process_response_body_buffer(buffer)
 
-    assert mock_manager.invoke_hook.called
-    payload = mock_manager.invoke_hook.call_args[0][1]
-    verify_payload_content(payload, tool_result["result"], "Single chunk")
-    # Verify ProcessingResponse was returned
+    # Verify hook is NOT called for error responses
+    assert not mock_manager.invoke_hook.called, (
+        "Tool post-invoke hook should not be called for error responses"
+    )
+    # Verify response is returned (function handles error responses gracefully)
     assert response is not None

From 5eaa94c3abc01fd373a09d722963e3224faafa70 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Mon, 16 Feb 2026 10:36:33 -0700
Subject: [PATCH 24/27] :goal_net: Update immediate response for not continued
 processing

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py        | 65 ++++++++++++++++++++++++++------------------
 tests/test_server.py | 42 ++++++++++++++++++++++++----
 2 files changed, 76 insertions(+), 31 deletions(-)

diff --git a/src/server.py b/src/server.py
index 2df83be..4848ea7 100644
--- a/src/server.py
+++ b/src/server.py
@@ -131,6 +131,10 @@ async def getToolPostInvokeResponse(body):
 
     Invokes plugins after a tool has been called, allowing for result validation,
     modification, or filtering of the tool output.
+
+    Note: In STREAMED mode, blocking responses may fail if headers are already sent.
+    This implementation uses immediate_response to attempt early termination, but
+    it may not always succeed due to streaming constraints.
     """
     # FIXME: size of content array is expected to be 1
     # for content in body["result"]["content"]:
@@ -145,37 +149,45 @@ async def getToolPostInvokeResponse(body):
     )
     logger.debug(f"**** Tool Post Invoke result {result}")
     if not result.continue_processing:
-        # Build error response to replace the tool result
+        # Build error message
+        error_message = "Tool response forbidden"
+        if result.violation is not None:
+            violation: PluginViolation = result.violation
+            error_message = f"{violation.reason} -- {violation.description}"
+
         error_body = {
             "jsonrpc": body["jsonrpc"],
             "id": body["id"],
-            "error": {"code": -32000, "message": "Tool response forbidden"},
+            "error": {"code": -32000, "message": error_message},
         }
+
+        # In STREAMED mode, we attempt to use immediate_response to terminate early
+        # This may fail if response headers have already been sent
+        logger.warning(
+            "Tool post-invoke blocking in STREAMED mode - may be unreliable. "
+            "Consider using BUFFERED mode or moving validation to pre-invoke hooks."
+        )
         body_resp = ep.ProcessingResponse(
-            response_body=ep.BodyResponse(
-                response=ep.CommonResponse(
-                    header_mutation=ep.HeaderMutation(
-                        set_headers=[
-                            core.HeaderValueOption(
-                                header=core.HeaderValue(
-                                    key="content-type",
-                                    raw_value="application/json".encode(
-                                        "utf-8"
-                                    ),
-                                )
-                            ),
-                            core.HeaderValueOption(
-                                header=core.HeaderValue(
-                                    key="x-mcp-denied",
-                                    raw_value="True".encode("utf-8"),
-                                )
-                            ),
-                        ],
-                    ),
-                    body_mutation=ep.BodyMutation(
-                        body=(json.dumps(error_body)).encode("utf-8")
-                    ),
-                )
+            immediate_response=ep.ImmediateResponse(
+                # Use 200 status with error in body for MCP protocol compatibility
+                status=http_status_pb2.HttpStatus(code=200),
+                headers=ep.HeaderMutation(
+                    set_headers=[
+                        core.HeaderValueOption(
+                            header=core.HeaderValue(
+                                key="content-type",
+                                raw_value="application/json".encode("utf-8"),
+                            )
+                        ),
+                        core.HeaderValueOption(
+                            header=core.HeaderValue(
+                                key="x-mcp-denied",
+                                raw_value="True".encode("utf-8"),
+                            )
+                        ),
+                    ],
+                ),
+                body=(json.dumps(error_body)).encode("utf-8"),
             )
         )
     else:
@@ -192,6 +204,7 @@ async def getToolPostInvokeResponse(body):
         else:
             body_mutation = ep.BodyResponse(response=ep.CommonResponse())
         body_resp = ep.ProcessingResponse(response_body=body_mutation)
+    logger.info(f"****Tool Post Invoke Return body: {body_resp}****")
     return body_resp
 
 
diff --git a/tests/test_server.py b/tests/test_server.py
index 64dcfe6..ab425bb 100644
--- a/tests/test_server.py
+++ b/tests/test_server.py
@@ -147,7 +147,14 @@ async def test_getToolPostInvokeResponse_continue_processing(
 async def test_getToolPostInvokeResponse_blocked(
     mock_envoy_modules, mock_manager, sample_tool_result_body
 ):
-    """Test getToolPostInvokeResponse when plugin blocks the response."""
+    """Test getToolPostInvokeResponse when plugin blocks the response.
+
+    This test verifies that when continue_processing=False, the function
+    uses immediate_response (not response_body) and includes violation details.
+    """
+    # Setup mocks for immediate_response path
+    setup_response_mocks(mock_envoy_modules)
+
     # Import server after mocking
     import src.server
 
@@ -166,10 +173,23 @@ async def test_getToolPostInvokeResponse_blocked(
     # Inject mock manager
     src.server.manager = mock_manager
 
-    # Call the function
-    response = await src.server.getToolPostInvokeResponse(
-        sample_tool_result_body
-    )
+    # Capture json.dumps calls to verify error body content
+    original_dumps = json.dumps
+    captured_bodies = []
+
+    def spy_dumps(obj, **kwargs):
+        if isinstance(obj, dict) and "error" in obj:
+            captured_bodies.append(obj)
+        return original_dumps(obj, **kwargs)
+
+    json.dumps = spy_dumps
+    try:
+        # Call the function
+        response = await src.server.getToolPostInvokeResponse(
+            sample_tool_result_body
+        )
+    finally:
+        json.dumps = original_dumps
 
     # Verify the hook was called with correct payload
     assert mock_manager.invoke_hook.called
@@ -181,6 +201,18 @@ async def test_getToolPostInvokeResponse_blocked(
     # Verify response was created (error path taken)
     assert response is not None
 
+    # Verify error body was created with violation details
+    assert len(captured_bodies) > 0
+    error_body = captured_bodies[0]
+    assert "error" in error_body
+    assert error_body["error"]["code"] == -32000
+    # Verify violation message is included
+    assert "Sensitive content detected" in error_body["error"]["message"]
+    assert (
+        "Tool response contains forbidden content"
+        in error_body["error"]["message"]
+    )
+
 
 @pytest.mark.asyncio
 async def test_getToolPostInvokeResponse_modified_payload(

From 3fc2672a02c6602857c108aeb05dc725fd1729e5 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Mon, 16 Feb 2026 10:41:55 -0700
Subject: [PATCH 25/27] :recycle: Refactor common immediate response

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 139 ++++++++++++++++++++++----------------------------
 1 file changed, 62 insertions(+), 77 deletions(-)

diff --git a/src/server.py b/src/server.py
index 4848ea7..7db6b70 100644
--- a/src/server.py
+++ b/src/server.py
@@ -18,7 +18,6 @@
     PromptPrehookPayload,
     ToolPostInvokePayload,
     ToolPreInvokePayload,
-    PluginViolation,
 )
 
 from mcpgateway.plugins.framework import PluginManager
@@ -44,6 +43,56 @@ def set_result_in_body(body, result_args):
     body["params"]["arguments"] = result_args
 
 
+def create_mcp_immediate_error_response(body, error_message, violation=None):
+    """
+    Create an MCP error response using immediate_response.
+
+    This helper creates a standardized error response that can be used
+    for both pre-invoke and post-invoke blocking scenarios.
+
+    Args:
+        body: The original request/response body containing jsonrpc and id
+        error_message: Base error message
+        violation: Optional PluginViolation with reason and description
+
+    Returns:
+        ProcessingResponse with immediate_response containing the error
+    """
+    # Build error message with violation details if present
+    if violation is not None:
+        error_message = f"{violation.reason} -- {violation.description}"
+
+    error_body = {
+        "jsonrpc": body["jsonrpc"],
+        "id": body["id"],
+        "error": {"code": -32000, "message": error_message},
+    }
+
+    return ep.ProcessingResponse(
+        immediate_response=ep.ImmediateResponse(
+            # Use 200 status with error in body for MCP protocol compatibility
+            status=http_status_pb2.HttpStatus(code=200),
+            headers=ep.HeaderMutation(
+                set_headers=[
+                    core.HeaderValueOption(
+                        header=core.HeaderValue(
+                            key="content-type",
+                            raw_value="application/json".encode("utf-8"),
+                        )
+                    ),
+                    core.HeaderValueOption(
+                        header=core.HeaderValue(
+                            key="x-mcp-denied",
+                            raw_value="True".encode("utf-8"),
+                        )
+                    ),
+                ],
+            ),
+            body=(json.dumps(error_body)).encode("utf-8"),
+        )
+    )
+
+
 # ============================================================================
 # MCP HOOK HANDLERS
 # ============================================================================
@@ -73,37 +122,10 @@ async def getToolPreInvokeResponse(body):
     )
     logger.debug(f"**** Tool Pre Invoke Result: {result} ****")
     if not result.continue_processing:
-        error_message = "No go - Tool args forbidden"
-        if result.violation is not None:
-            violation: PluginViolation = result.violation
-            error_message = f"{violation.reason} -- {violation.description}"
-        error_body = {
-            "jsonrpc": body["jsonrpc"],
-            "id": body["id"],
-            "error": {"code": -32000, "message": error_message},
-        }
-        body_resp = ep.ProcessingResponse(
-            immediate_response=ep.ImmediateResponse(
-                # ok for stream, with error in body
-                status=http_status_pb2.HttpStatus(code=200),
-                headers=ep.HeaderMutation(
-                    set_headers=[
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="content-type",
-                                raw_value="application/json".encode("utf-8"),
-                            )
-                        ),
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="x-mcp-denied",
-                                raw_value="True".encode("utf-8"),
-                            )
-                        ),
-                    ],
-                ),
-                body=(json.dumps(error_body)).encode("utf-8"),
-            )
+        body_resp = create_mcp_immediate_error_response(
+            body,
+            error_message="No go - Tool args forbidden",
+            violation=result.violation,
         )
     else:
         logger.debug("continue_processing true")
@@ -149,46 +171,12 @@ async def getToolPostInvokeResponse(body):
     )
     logger.debug(f"**** Tool Post Invoke result {result}")
     if not result.continue_processing:
-        # Build error message
-        error_message = "Tool response forbidden"
-        if result.violation is not None:
-            violation: PluginViolation = result.violation
-            error_message = f"{violation.reason} -- {violation.description}"
-
-        error_body = {
-            "jsonrpc": body["jsonrpc"],
-            "id": body["id"],
-            "error": {"code": -32000, "message": error_message},
-        }
-
         # In STREAMED mode, we attempt to use immediate_response to terminate early
         # This may fail if response headers have already been sent
-        logger.warning(
-            "Tool post-invoke blocking in STREAMED mode - may be unreliable. "
-            "Consider using BUFFERED mode or moving validation to pre-invoke hooks."
-        )
-        body_resp = ep.ProcessingResponse(
-            immediate_response=ep.ImmediateResponse(
-                # Use 200 status with error in body for MCP protocol compatibility
-                status=http_status_pb2.HttpStatus(code=200),
-                headers=ep.HeaderMutation(
-                    set_headers=[
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="content-type",
-                                raw_value="application/json".encode("utf-8"),
-                            )
-                        ),
-                        core.HeaderValueOption(
-                            header=core.HeaderValue(
-                                key="x-mcp-denied",
-                                raw_value="True".encode("utf-8"),
-                            )
-                        ),
-                    ],
-                ),
-                body=(json.dumps(error_body)).encode("utf-8"),
-            )
+        body_resp = create_mcp_immediate_error_response(
+            body,
+            error_message="Tool response forbidden",
+            violation=result.violation,
         )
     else:
         result_payload = result.modified_payload
@@ -225,13 +213,10 @@ async def getPromptPreFetchResponse(body):
     )
     logger.info(result)
     if not result.continue_processing:
-        body_resp = ep.ProcessingResponse(
-            immediate_response=ep.ImmediateResponse(
-                status=http_status_pb2.HttpStatus(
-                    code=http_status_pb2.Forbidden
-                ),
-                details="No go",
-            )
+        body_resp = create_mcp_immediate_error_response(
+            body,
+            error_message="Tool response forbidden",
+            violation=result.violation,
         )
     else:
         body["params"]["arguments"] = result.modified_payload.args

From b9bf557a478b435a8f68bbd10c028853f73db179 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Fri, 20 Feb 2026 11:20:50 -0700
Subject: [PATCH 26/27] :recycle::goal_net: Consistent error responses

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 plugins/examples/nemocheck/plugin.py          | 36 ++++++-----
 .../nemocheck/tests/test_nemocheck.py         | 60 ++++++++++++++-----
 2 files changed, 66 insertions(+), 30 deletions(-)

diff --git a/plugins/examples/nemocheck/plugin.py b/plugins/examples/nemocheck/plugin.py
index cc87980..8d1f704 100644
--- a/plugins/examples/nemocheck/plugin.py
+++ b/plugins/examples/nemocheck/plugin.py
@@ -153,7 +153,7 @@ async def tool_pre_invoke(
                     violation = PluginViolation(
                         reason=f"Check tool rails:{status}.",
                         description=json.dumps(data),
-                        code=f"checkserver_http_status_code:{response.status_code}",
+                        code="NEMO_RAILS_BLOCKED",
                         details=metadata,
                     )
                     return ToolPreInvokeResult(
@@ -164,9 +164,9 @@ async def tool_pre_invoke(
             else:
                 violation = PluginViolation(
                     reason="Tool Check Unavailable",
-                    description="Tool arguments check server returned error",
-                    code=f"checkserver_http_status_code:{response.status_code}",
-                    details={},
+                    description=f"Tool arguments check server returned error. Status code: {response.status_code}, Response: {response.text}",
+                    code="NEMO_SERVER_ERROR",
+                    details={"status_code": response.status_code},
                 )
                 return ToolPreInvokeResult(
                     continue_processing=False, violation=violation
@@ -177,8 +177,8 @@ async def tool_pre_invoke(
             violation = PluginViolation(
                 reason="Tool Check Error",
                 description=f"Failed to connect to check server: {str(e)}",
-                code="checkserver_connection_error",
-                details={},
+                code="NEMO_CONNECTION_ERROR",
+                details={"error": str(e)},
             )
             return ToolPreInvokeResult(
                 continue_processing=False, violation=violation
@@ -248,9 +248,9 @@ async def tool_post_invoke(
                 else:  # blocked
                     metadata = data.get("rails_status")
                     violation = PluginViolation(
-                        reason=f"Tool response check status: {status}",
-                        description="Rails check blocked tool response",
-                        code=f"checkserver_http_status_code:{response.status_code}",
+                        reason=f"Check tool rails:{status}.",
+                        description=json.dumps(data),
+                        code="NEMO_RAILS_BLOCKED",
                         details=metadata,
                     )
                     result = ToolPostInvokeResult(
@@ -260,10 +260,10 @@ async def tool_post_invoke(
                     )
             else:
                 violation = PluginViolation(
-                    reason="Tool response check unavailable",
-                    description="Tool response check server returned error",
-                    code=f"checkserver_http_status_code:{response.status_code}",
-                    details={},
+                    reason="Tool Check Unavailable",
+                    description=f"Tool response check server returned error. Status code: {response.status_code}, Response: {response.text}",
+                    code="NEMO_SERVER_ERROR",
+                    details={"status_code": response.status_code},
                 )
                 result = ToolPostInvokeResult(
                     continue_processing=False, violation=violation
@@ -274,6 +274,12 @@ async def tool_post_invoke(
 
         except Exception as e:
             logger.error(f"[NemoCheck] Error checking tool response: {e}")
+            violation = PluginViolation(
+                reason="Tool Check Error",
+                description=f"Failed to connect to check server: {str(e)}",
+                code="NEMO_CONNECTION_ERROR",
+                details={"error": str(e)},
+            )
             return ToolPostInvokeResult(
-                continue_processing=True
-            )  # Fail open on error
+                continue_processing=False, violation=violation
+            )
diff --git a/plugins/examples/nemocheck/tests/test_nemocheck.py b/plugins/examples/nemocheck/tests/test_nemocheck.py
index 9fe24c1..7c1cbd7 100644
--- a/plugins/examples/nemocheck/tests/test_nemocheck.py
+++ b/plugins/examples/nemocheck/tests/test_nemocheck.py
@@ -59,7 +59,7 @@ async def test_prompt_pre_fetch(plugin, context):
 
 @pytest.mark.asyncio
 @pytest.mark.parametrize(
-    "status_code,response_data,expected_continue,has_violation",
+    "status_code,response_data,expected_continue,has_violation,expected_code",
     [
         (
             200,
@@ -71,6 +71,7 @@ async def test_prompt_pre_fetch(plugin, context):
             },
             True,
             False,
+            None,
         ),
         (
             200,
@@ -80,8 +81,9 @@ async def test_prompt_pre_fetch(plugin, context):
             },
             False,
             True,
+            "NEMO_RAILS_BLOCKED",
         ),
-        (503, None, False, True),
+        (503, None, False, True, "NEMO_SERVER_ERROR"),
     ],
 )
 async def test_tool_pre_invoke_scenarios(
@@ -91,8 +93,9 @@ async def test_tool_pre_invoke_scenarios(
     response_data,
     expected_continue,
     has_violation,
+    expected_code,
 ):
-    """Test tool_pre_invoke with various scenarios."""
+    """Test tool_pre_invoke with various scenarios including error codes."""
     payload = ToolPreInvokePayload(
         name="test_tool",
         args={"tool_args": '{"param": "value"}'},
@@ -106,11 +109,13 @@ async def test_tool_pre_invoke_scenarios(
 
     assert result.continue_processing == expected_continue
     assert (result.violation is not None) == has_violation
+    if has_violation:
+        assert result.violation.code == expected_code
 
 
 @pytest.mark.asyncio
 @pytest.mark.parametrize(
-    "status_code,response_data,expected_continue,has_violation",
+    "status_code,response_data,expected_continue,has_violation,expected_code",
     [
         (
             200,
@@ -122,6 +127,7 @@ async def test_tool_pre_invoke_scenarios(
             },
             True,
             False,
+            None,
         ),
         (
             200,
@@ -131,8 +137,9 @@ async def test_tool_pre_invoke_scenarios(
             },
             False,
             True,
+            "NEMO_RAILS_BLOCKED",
         ),
-        (500, None, False, True),
+        (500, None, False, True, "NEMO_SERVER_ERROR"),
     ],
 )
 async def test_tool_post_invoke_http_scenarios(
@@ -142,8 +149,9 @@ async def test_tool_post_invoke_http_scenarios(
     response_data,
     expected_continue,
     has_violation,
+    expected_code,
 ):
-    """Test tool_post_invoke with various HTTP response scenarios."""
+    """Test tool_post_invoke with various HTTP response scenarios including error codes."""
     payload = ToolPostInvokePayload(
         name="test_tool",
         result={"content": [{"type": "text", "text": "Test content"}]},
@@ -157,6 +165,8 @@ async def test_tool_post_invoke_http_scenarios(
 
     assert result.continue_processing == expected_continue
     assert (result.violation is not None) == has_violation
+    if has_violation:
+        assert result.violation.code == expected_code
 
 
 @pytest.mark.asyncio
@@ -230,15 +240,35 @@ async def test_tool_post_invoke_filters_non_text(plugin, context):
 
 
 @pytest.mark.asyncio
-async def test_tool_post_invoke_fails_open_on_exception(plugin, context):
-    """Test tool_post_invoke fails open on exceptions."""
-    payload = ToolPostInvokePayload(
-        name="test_tool",
-        result={"content": [{"type": "text", "text": "content"}]},
-    )
+@pytest.mark.parametrize(
+    "hook_name,payload_factory",
+    [
+        (
+            "tool_pre_invoke",
+            lambda: ToolPreInvokePayload(
+                name="test_tool", args={"tool_args": '{"param": "value"}'}
+            ),
+        ),
+        (
+            "tool_post_invoke",
+            lambda: ToolPostInvokePayload(
+                name="test_tool",
+                result={"content": [{"type": "text", "text": "content"}]},
+            ),
+        ),
+    ],
+)
+async def test_connection_error_handling(
+    plugin, context, hook_name, payload_factory
+):
+    """Test both hooks fail closed on connection errors with NEMO_CONNECTION_ERROR code."""
+    payload = payload_factory()
+    hook = getattr(plugin, hook_name)
 
     with patch("plugin.requests.post", side_effect=Exception("Network error")):
-        result = await plugin.tool_post_invoke(payload, context)
+        result = await hook(payload, context)
 
-    assert result.continue_processing
-    assert result.violation is None
+    assert not result.continue_processing
+    assert result.violation is not None
+    assert result.violation.code == "NEMO_CONNECTION_ERROR"
+    assert "Network error" in result.violation.description

From 946237c511348e11c25e95bc571939cd7a2c2934 Mon Sep 17 00:00:00 2001
From: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
Date: Mon, 23 Feb 2026 12:06:22 -0700
Subject: [PATCH 27/27] :recycle: Refactor post tool invoke processing

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>
---
 src/server.py | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)

diff --git a/src/server.py b/src/server.py
index 7db6b70..e7785e5 100644
--- a/src/server.py
+++ b/src/server.py
@@ -178,20 +178,23 @@ async def getToolPostInvokeResponse(body):
             error_message="Tool response forbidden",
             violation=result.violation,
         )
-    else:
-        result_payload = result.modified_payload
-        if result_payload is not None:
-            body["result"] = result_payload.result
-            body_mutation = ep.BodyResponse(
-                response=ep.CommonResponse(
-                    body_mutation=ep.BodyMutation(
-                        body=json.dumps(body).encode("utf-8")
-                    )
+        logger.info(f"****Tool Post Invoke Return body: {body_resp}****")
+        return body_resp
+
+    # Continue processing - allow or modify the response
+    result_payload = result.modified_payload
+    if result_payload is not None:
+        body["result"] = result_payload.result
+        body_mutation = ep.BodyResponse(
+            response=ep.CommonResponse(
+                body_mutation=ep.BodyMutation(
+                    body=json.dumps(body).encode("utf-8")
                 )
             )
-        else:
-            body_mutation = ep.BodyResponse(response=ep.CommonResponse())
-        body_resp = ep.ProcessingResponse(response_body=body_mutation)
+        )
+    else:
+        body_mutation = ep.BodyResponse(response=ep.CommonResponse())
+    body_resp = ep.ProcessingResponse(response_body=body_mutation)
     logger.info(f"****Tool Post Invoke Return body: {body_resp}****")
     return body_resp