ardea

MCP tools for AI agents to report API experiences. Record observations during tasks and submit structured feedback that gets scored and surfaced in the Canary dashboard.

Install

One command to add to Claude Code:

claude mcp add ardea -- npx -y ardea

That's it. On first run a browser window opens for signup — your API key is saved automatically.

Other MCP clients

Add to your .mcp.json (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "ardea": {
      "command": "npx",
      "args": ["-y", "ardea"]
    }
  }
}

Tools

ardea_annotate

Record API observations during a task. Stores notes locally in ~/.ardea/annotations/.

• id — annotation key, e.g. "stripe/charges"
• note — what happened: result, errors, latency
• list: true — list all saved annotations
• clear: true — delete an annotation

ardea_report

Submit a structured API experience report for scoring.

• task_goal (required) — what you were trying to do
• outcome (required) — "success", "partial", "failure", or "blocked"
• apis_used (required) — APIs and endpoints used
• what_worked — what went well
• what_failed — what broke, with status codes and errors
• friction — biggest pain point
• suggestion — improvement recommendation

Environment Variables

All optional — browser signup handles configuration automatically.

| Variable | Purpose | |----------|---------| | ARDEA_API_KEY | Skip browser signup | | ARDEA_ENDPOINT | Custom backend URL (default: production) | | ARDEA_SESSION_ID | Correlate reports to a session | | ARDEA_AGENT_NAME | Label for this agent |

License

MIT