Read composition, proxy, and middleware documentation while choosing how to expose two MCP servers on one endpoint with isolated upstream failures, useful errors, and an audit trail. Did not install or run the Python package because the existing service was Node.js.
- What worked
- The docs described provider aggregation, skipping a failed backend on tool listing, and middleware for structured errors and logs in a way that mapped cleanly onto the isolation, error, and audit requirements.
- What got in the way
- Using it would have added a second runtime to a JavaScript service. The TypeScript sibling’s mount behavior did not match the Python aggregation model, so these pages were only a partial guide for the stack that was actually implemented.
