Prometheus MCP Server

What is Prometheus MCP Server?

Prometheus MCP Server is a Model Context Protocol server that enables LLMs to interact with a running Prometheus instance via its HTTP API. It allows generating and executing PromQL queries, listing and analyzing metrics, and reading Prometheus documentation, targeting developers and SREs who want AI-driven observability.

How to use Prometheus MCP Server?

Install the binary and configure it to connect to a Prometheus instance. Use command-line flags such as --prometheus.url (default: http://localhost:9090), --mcp.tools to whitelist specific tools, --dangerous.enable-tsdb-admin-tools for destructive operations, --prometheus.backend for compatible backends, --toon for TOON encoding, and --truncation to limit API response lines. The server exposes tools and resources that LLM clients can call.

Key features of Prometheus MCP Server

• Execute instant and range PromQL queries
• List and search metrics, labels, and series
• Access official Prometheus documentation via docs tools
• Support for Prometheus compatible backends (e.g., Thanos)
• Token/context efficiency through TOON encoding and response truncation
• Optional TSDB admin tools (dangerous)

Use cases of Prometheus MCP Server

• Investigate slow queries by analyzing Prometheus metrics
• Suggest recording rules for SLOs based on server metrics
• Summarize Prometheus metric and label naming best practices
• Review Prometheus instance health, configuration, and runtime info

FAQ from Prometheus MCP Server

What Prometheus compatible backends are supported?

The server works with any service offering a Prometheus compatible API. Specific backends like Thanos can be selected via --prometheus.backend, which may add or remove tools (e.g., Thanos removes config and adds list_stores).

Can I limit token usage from Prometheus API responses?

Yes. Use the --toon flag to convert JSON to TOON format for token efficiency, or set --truncation to limit the number of lines/entries returned. The truncation limit is disabled by default and can be overridden per-tool-call by capable LLMs.

What are the runtime dependencies?

The server is a compiled Go binary with no external runtime dependencies. It requires network access to a running Prometheus instance (or compatible backend).

How do I enable TSDB admin tools?

TSDB admin endpoints (e.g., delete_series, snapshot, clean_tombstones) are disabled by default. Start the server with the flag --dangerous.enable-tsdb-admin-tools to acknowledge the risk and enable them.

Does the server support authentication or transport security?

The README does not mention authentication or TLS configuration. The server assumes a direct HTTP connection to Prometheus and relies on Prometheus's own access control.