From d34d9c6986e621163ff69ae1839429f2ac00d381 Mon Sep 17 00:00:00 2001
From: Jakub Hrozek <jakub@stacklok.com>
Date: Fri, 17 Jul 2026 18:11:06 +0100
Subject: [PATCH] Capture proxy logs in conformance CI runs

The MCP conformance job fails intermittently on tools-call-sampling with a
60s "Request timed out" on the client side. The uploaded results artifact
only contains the conformance client's view (checks.json), which cannot tell
us whether the sampling server->client round-trip stalled in the ToolHive
proxy or in the reference server, so the flake has been undiagnosable from
CI alone and does not reproduce locally.

Capture the ToolHive proxy logs (thv logs --proxy) into the results directory
on teardown, tee the suite output alongside them, and run the proxied server
with --debug so the proxy logs show the per-request message flow. The workflow
already uploads test/conformance/results on always(), so these land in the
existing conformance-results artifact with no workflow change.

This adds no fix; it makes the next CI failure debuggable.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 test/conformance/run-conformance.sh | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/test/conformance/run-conformance.sh b/test/conformance/run-conformance.sh
index 5749fc6cd9..9b679f34ae 100755
--- a/test/conformance/run-conformance.sh
+++ b/test/conformance/run-conformance.sh
@@ -30,6 +30,11 @@ export TOOLHIVE_SKIP_DESKTOP_CHECK=1
 
 CLONE_DIR=""
 cleanup() {
+  # Diagnostics: capture the ToolHive proxy logs before teardown so a CI failure
+  # (e.g. the sampling round-trip timing out) is debuggable from the uploaded
+  # results artifact, not just the client-side checks.json.
+  mkdir -p "${RESULTS_DIR}"
+  "${THV_BINARY}" logs --proxy "${SERVER_NAME}" > "${RESULTS_DIR}/proxy-logs.txt" 2>&1 || true
   "${THV_BINARY}" rm "${SERVER_NAME}" >/dev/null 2>&1 || true
   [ -n "${CLONE_DIR}" ] && rm -rf "${CLONE_DIR}" || true
 }
@@ -61,7 +66,7 @@ docker build -t "${IMAGE}" "${SERVER_DIR}"
 echo "==> Starting server through ToolHive"
 "${THV_BINARY}" rm -f "${SERVER_NAME}" >/dev/null 2>&1 || true
 "${THV_BINARY}" run "${IMAGE}" \
-  --transport streamable-http --target-port 3000 --name "${SERVER_NAME}"
+  --transport streamable-http --target-port 3000 --name "${SERVER_NAME}" --debug
 
 echo "==> Resolving proxy URL"
 URL=""
@@ -89,8 +94,9 @@ done
 
 echo "==> Running conformance suite (${CONFORMANCE_SUITE})"
 rm -rf "${RESULTS_DIR}"
+mkdir -p "${RESULTS_DIR}"
 npx -y "@modelcontextprotocol/conformance@${CONFORMANCE_VERSION}" server \
   --url "${URL}" \
   --suite "${CONFORMANCE_SUITE}" \
   --expected-failures "${EXPECTED_FAILURES}" \
-  -o "${RESULTS_DIR}"
+  -o "${RESULTS_DIR}" 2>&1 | tee "${RESULTS_DIR}/suite-output.log"