Real-world patterns for AI agents with persistent identity, memory, and coordination.
AgentID gives your AI agents a persistent URL, shared memory, and coordination primitives via a single MCP server. No backend required.
Your agent code
│
▼
AgentID MCP ──────────────────────────────────────────────────
│ write_memory read_memory search_memory
│ start_mission update_status read_mission
│ handoff report_activity get_secret
▼
Persistent store (memory, missions, activity feed, secret vault)
│
▼
Any other agent connected to the same persona reads it too
| Example | Primitives used | What it shows |
|---|---|---|
| persistent-memory | write_memory read_memory search_memory |
Hello world — agents that remember across sessions |
| code-reviewer | write_memory search_memory report_activity |
Single agent that builds codebase knowledge over time |
| daily-standup | write_memory search_memory |
Log work all day, generate standup in the morning |
| research-writer | write_memory handoff start_mission |
Two agents hand off work through shared memory |
| multi-agent-agency | handoff write_memory read_mission |
Three-agent pipeline: researcher → analyst → writer |
| mission-tracker | start_mission update_status read_mission |
PM + engineer agents coordinate via structured missions |
| api-key-vault | get_secret |
Agents use stored API keys without ever seeing them in config |
| team-workflow | write_memory read_memory handoff |
Two agents share one persona — one writes, one reads |
Prerequisites: Python 3.10+, an AgentID account (free).
# 1. Clone
git clone https://github.com/agentid-live/examples
cd examples
# 2. Install dependencies
pip install anthropic mcp
# 3. Set your API key (get it from agentid.live/app/developers)
export AGENTID_API_KEY=ak_live_...
# 4. Run any example
python persistent-memory/demo.pyEach example also works with Claude Desktop — copy the claude_desktop_config.json into your config and talk to the agent directly.
Every agent gets these tools via its MCP URL (https://agentid.live/api/mcp/{handle}).
# Persist anything under a named key
write_memory(key="user_prefs", value="timezone: UTC, language: Python")
# Retrieve by exact key
read_memory(key="user_prefs")
# Semantic search across all stored memory
search_memory(query="what timezone is the user in?")# Create a structured task
mission_id = start_mission(title="Analyze Q4 data", goal="Identify top 3 trends")
# Update progress
update_status(mission_id=mission_id, status="in_progress", notes="Processing CSV files")
# Read current state (any agent on the same persona sees this)
read_mission(mission_id=mission_id)# Pass work to another agent — it sees this on next read_mission() call
handoff(
to_agent="writer_agent",
context="Research complete. Memory keys: blog_outline, section_1, section_2.",
mission_id=mission_id,
)# Log an event to the activity feed (visible in the studio)
report_activity(summary="Reviewed PR #42: found 3 security issues", type="review")
# Retrieve an API key stored in AgentID Drive — never in your config
api_key = get_secret(name="posthog_api_key")All demos import from shared/agent_runner.py, which handles the MCP connection and Anthropic tool-use loop:
from shared.agent_runner import run_agent
result = await run_agent(
handle="my_agent",
system="You are @my_agent. ...",
user_message="Do something useful.",
)Add any agent to Claude Desktop in one step:
{
"mcpServers": {
"my-agent": {
"url": "https://agentid.live/api/mcp/my-agent",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}Works the same in Cursor, Windsurf, or any MCP-compatible client.
- agentid.live/showcase — see agents in action
- agentid.live/app/studio — inspect memory, missions, and activity feed
- Contributing — add your own example
- agentid.live/register — free account, no credit card