Query Prometheus Metrics with AI

Prometheus MCP Server gives AI assistants access to a Prometheus server's metrics and queries through standardized MCP interfaces, letting them execute PromQL queries and analyze metrics data directly. It exposes 6 tools: health_check (container health/status verification), execute_query (a PromQL instant query), execute_range_query (a PromQL range query with start time, end time, and step interval), list_metrics (lists available metrics with pagination and filtering), get_metric_metadata (metadata for one metric or bulk, with optional filtering), and get_targets (scrape targets, with server-side state/scrape_pool filtering and pagination). Which tools are exposed to the client is configurable, so you can trim the set to what you actually use and save context window.

Use it when you want an assistant to answer questions about your metrics directly - current values, historical trends over a time range, which metrics exist, or which scrape targets are up - without you hand-writing PromQL and pasting results back in. Basic auth, bearer-token auth, and mutual TLS (client certificate/key) are all supported for connecting to a secured Prometheus server, plus a PROMETHEUS_CUSTOM_HEADERS option and TOOL_PREFIX for running multiple instances against different environments (e.g. staging_execute_query) side by side in the same client.

Don't point it at a Prometheus instance you don't want an AI assistant querying at will - there's no scoping of which metrics or targets a tool can see beyond what the Prometheus server itself returns, and PROMETHEUS_REQUEST_TIMEOUT (default 30s) is the only built-in protection against a runaway or hanging query.

Beyond the core query and discovery tools, the server supports SSL verification toggling (PROMETHEUS_URL_SSL_VERIFY), a PROMETHEUS_DISABLE_LINKS flag to strip Prometheus UI links from results and save context tokens, multi-tenant ORG_ID headers, and three transport modes - stdio (default), HTTP, and SSE - configurable via PROMETHEUS_MCP_SERVER_TRANSPORT, with PROMETHEUS_MCP_STATELESS_HTTP available for multi-replica deployments.

For Claude Desktop or comparable clients, run the published container:

{
  "mcpServers": {
    "prometheus": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "PROMETHEUS_URL",
        "ghcr.io/pab1it0/prometheus-mcp-server:latest"
      ],
      "env": {
        "PROMETHEUS_URL": "<your-prometheus-url>"
      }
    }
  }
}

Claude Code installs via claude mcp add, and a Helm chart (oci://ghcr.io/pab1it0/charts/prometheus-mcp-server) is available for Kubernetes deployments with the same environment-variable configuration surface. Docker Desktop users can add it through the MCP Catalog or the MCP Toolkit extension.

SREs, platform engineers, and developers who want an AI assistant to query and reason over their existing Prometheus metrics - instant and range queries, metric discovery, and target health - instead of switching to the Prometheus UI or writing PromQL by hand for every question.

The project is released under the MIT License.