Prometheus MCP Server

Content

Prometheus MCP Server gives Claude and other MCP clients direct access to Prometheus-compatible metrics. It can run PromQL instant queries, run range queries over a time window, list metric names, inspect metric metadata, fetch scrape targets, and report server health.

Use it when an operations, SRE, or platform workflow needs Claude to inspect metrics without routing through Grafana. It supports Docker-based stdio setup, HTTP and SSE transport modes, Kubernetes deployment through Helm, optional auth, custom headers, request timeouts, and tool prefixes for multiple environments.

Source Review

• https://github.com/pab1it0/prometheus-mcp-server
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/README.md
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/LICENSE
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/pyproject.toml
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/server.json
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/src/prometheus_mcp_server/server.py
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/src/prometheus_mcp_server/main.py
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/charts/prometheus-mcp-server/values.yaml
• https://raw.githubusercontent.com/pab1it0/prometheus-mcp-server/main/Dockerfile

These sources were reviewed on 2026-06-06. Prefer the live repository, README, license, Python package metadata, server manifest, server implementation, entry point, Helm values, and Dockerfile for current setup and operating details.

Features

• Run PromQL instant queries with execute_query.
• Run PromQL range queries with start time, end time, and step interval.
• List available metric names with pagination and filtering.
• Fetch metric metadata for one metric or bulk metadata.
• Inspect scrape targets and their health.
• Report MCP server and Prometheus connectivity health.
• Configure basic auth, bearer token auth, mutual TLS, custom headers, and organization IDs for multi-tenant setups.
• Choose stdio, HTTP, or SSE MCP transport.
• Set request timeouts to limit hanging or expensive queries.
• Disable Prometheus UI links in responses to reduce context size and avoid leaking internal URLs.
• Prefix tool names so multiple instances can target different environments.
• Deploy with Docker, Docker Desktop MCP Toolkit, or the bundled Helm chart.

Installation

Run the GHCR container with a single Prometheus endpoint:

docker run -i --rm \
  -e PROMETHEUS_URL="https://prometheus.example.com" \
  ghcr.io/pab1it0/prometheus-mcp-server:latest

Add it to Claude Desktop:

{
  "mcpServers": {
    "prometheus": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "PROMETHEUS_URL",
        "ghcr.io/pab1it0/prometheus-mcp-server:latest"
      ],
      "env": {
        "PROMETHEUS_URL": "https://prometheus.example.com"
      }
    }
  }
}

For a protected Prometheus endpoint:

docker run -i --rm \
  -e PROMETHEUS_URL="https://prometheus.example.com" \
  -e PROMETHEUS_TOKEN="YOUR_TOKEN" \
  ghcr.io/pab1it0/prometheus-mcp-server:latest

Deploy to Kubernetes with Helm:

helm install prometheus-mcp-server \
  oci://ghcr.io/pab1it0/charts/prometheus-mcp-server \
  --set prometheus.url="https://prometheus.example.com"

Use Cases

• Ask Claude which services have elevated error rates or latency over a time window.
• Compare deployment, namespace, or tenant metrics during an incident review.
• Discover available metrics and metadata before writing dashboards or alerts.
• Inspect scrape target health and Prometheus connectivity.
• Run read-only SRE investigations from an MCP client without opening broad shell access.
• Use tool prefixes to connect separate staging and production Prometheus instances to the same MCP client.

Safety and Privacy

Prometheus data often describes live infrastructure. Even read-only PromQL can surface sensitive service names, customers, traffic patterns, hostnames, incident details, and internal URLs. Limit the configured endpoint, prefer least-privilege credentials, use request timeouts, and disable UI links when internal Prometheus URLs should not enter the model context.

Large or repeated range queries can create real operational load. Treat Claude's Prometheus access as production observability access: keep credentials secret, avoid broad autonomous loops, and verify incident-response conclusions before acting on them.

Duplicate Notes

The catalog already includes Grafana MCP coverage, including querying Prometheus through Grafana datasources. This entry covers the separate pab1it0/prometheus-mcp-server project, which connects directly to a Prometheus-compatible API and ships Docker, GHCR, and Helm deployment paths.