mulewatch plugs your AI assistant into MuleSoft Anypoint Platform so you can ask about a Mule estate in plain English and get answers from the real platform APIs. It's built for the person holding the pager, not the person writing the flow.
It does the one thing the official MuleSoft MCP server doesn't: read and search application logs, including the historical ones in the Monitoring Archive.
What you can ask it
- Latest logs of an app, normalized across CloudHub 2.0, Runtime Fabric and legacy CloudHub.
- Just the
ERRORandFATALlines, with counts by logger and replica. - Logs from a past date, pulled from the Anypoint Monitoring Archive.
- What's deployed where, which API Manager instances front it, and what's in Exchange.
Every tool is read-only. There's no deploy, restart or policy tool, on purpose.
The interesting bit
The Monitoring Archive is indexed per replica, not per app, so a naive search has to probe every replica an app ever had. mulewatch tries the replica that's running right now first, and only falls back to a full scan when the app was redeployed since. On a test app with 669 archived replicas, a search for today probed exactly one.
I wrote the whole story up here: The official MuleSoft MCP server can't read your logs, so I built one that can.
Run it
claude mcp add mulewatch --env ANYPOINT_CLIENT_ID=... --env ANYPOINT_CLIENT_SECRET=... -- npx -y mulewatchIt works the same in Claude Desktop, Cursor and OpenClaw with an npx -y mulewatch entry. Lock it to safe environments with ANYPOINT_ALLOWED_ENVIRONMENTS=Dev,Sandbox.
Plain text: mulewatch.md · All projects