Hey team. fastapi_mcp user here. Wanted to flag MCPWatch in case it's useful to surface in your docs.
The pattern: instrument(mcp) wraps any FastMCP server (including fastapi_mcp) and gives you per-tool latency, error rates, p50/p95/p99 percentiles, and silent-failure detection. One line of code, no config.
For your users running fastapi_mcp in production, this fills the observability gap. Tools like Sentry/Datadog don't know how to instrument MCP. They see HTTP calls but lose the per-tool semantics.
If you'd consider a docs example or examples/ folder addition, I'm happy to send a tested PR.
Article with full breakdown: https://dev.to/alexlaguardia/your-mcp-servers-are-flying-blind-heres-how-to-fix-it-3g51
Repo: https://github.com/AlexlaGuardia/Vigil
No pressure if it's out of scope. Wanted to flag the natural fit.
Hey team. fastapi_mcp user here. Wanted to flag MCPWatch in case it's useful to surface in your docs.
The pattern:
instrument(mcp)wraps any FastMCP server (including fastapi_mcp) and gives you per-tool latency, error rates, p50/p95/p99 percentiles, and silent-failure detection. One line of code, no config.For your users running fastapi_mcp in production, this fills the observability gap. Tools like Sentry/Datadog don't know how to instrument MCP. They see HTTP calls but lose the per-tool semantics.
If you'd consider a docs example or examples/ folder addition, I'm happy to send a tested PR.
Article with full breakdown: https://dev.to/alexlaguardia/your-mcp-servers-are-flying-blind-heres-how-to-fix-it-3g51
Repo: https://github.com/AlexlaGuardia/Vigil
No pressure if it's out of scope. Wanted to flag the natural fit.