Enable MCP server for tempo

[SecretiveShell]

It would be really useful to enable and expose the embedded MCP server in tempo as part of the default configuration.

https://grafana.com/docs/tempo/latest/api_docs/mcp-server/#configuration

Currently I run LGTM locally during development to read logs/traces, and having the MCP server enabled would enable my AI assistant to take advantage of this extra information when debugging. Right now I have to rely on codex using the bash tool to run docker compose logs <service>, but this does not expose as much useful information as raw traces.

[martincostello]

Sounds reasonable for us to enable this (or make it easy to opt-in to) if it's a component included within Tempo.

Would you like to take a look at implementing this?

[zeitlinger]

+1, this fits well with the project's goal of making local development easy. As Martin said, PRs welcome!

Here's the MCP landscape for the components in this stack:

• Tempo: built-in MCP server (docs) — this issue
• Loki: grafana/loki-mcp — standalone server
• Grafana + Prometheus + Pyroscope: grafana/mcp-grafana — single server covering dashboards, alerts, and datasource queries
• OTel Collector: no MCP server yet (no issues/PRs in upstream repos)
• OBI: no MCP server yet (no issues/PRs in upstream repos)

We could expand this issue to cover wiring up all the mature ones (Tempo, Loki, mcp-grafana). That would give AI assistants access to traces, logs, metrics, profiles, and dashboards out of the box.

[zeitlinger]

Filed open-telemetry/opentelemetry-ebpf-instrumentation#1449 for OBI MCP server support.