Part of Grafana, not a vendor of its own. The four checks, setup and context cost live in the Grafana guide.
Part of the Grafana stack
Grafana's trace store. It is queried with TraceQL, and unlike the rest of the stack it runs an MCP server of its own, built into the Tempo query frontend.
Two routes, and they are not equivalent. The main Grafana connector has no TraceQL tool at all, though one tool, find_slow_requests, does search Tempo datasources for slow requests. Everything else needs Tempo's own MCP server, enabled with query_frontend.mcp_server.enabled: true and served at /api/mcp over streamable HTTP.
traceql-searchget-tracetraceql-metrics-instanttraceql-metrics-rangeget-attribute-namesget-attribute-valuesdocs-traceqldocs-configEight tools, from Tempo's own MCP server documentation rather than from the Grafana capture, because this server is separate. It also publishes six documentation resources under docs://, four for TraceQL and two for configuration.
The main connector cannot answer a trace question. It has no TraceQL tool, so an agent asked about a trace either answers from logs and metrics instead, or says it cannot answer, unless Tempo's own server is connected alongside it.
What to do: Connect both, and expect the trace question to route to the second. Tempo's documentation warns that enabling the server may pass tracing data to an LLM provider, which is a decision to make before turning it on.
find_slow_requests searches Tempo datasourcesquery_frontend.mcp_server.enabled: true--query-frontend.mcp-server.enabled=true/api/mcp on the query frontendGrafana Tempo is part of Grafana, not a vendor of its own
The four checks, the setup, the context cost and the read-only configuration are all decided at the Grafana level, so they live in that guide rather than being repeated here.