MCP Prometheus

MCP Prometheus is an MCP server that exposes Prometheus and Mimir time-series databases to AI assistants through standardized interfaces. Written in Go by Giant Swarm, it is primarily designed for platform engineers, DevOps teams, and site reliability engineers who need AI clients to inspect infrastructure health, investigate alerts, and execute queries. The server connects directly to the Prometheus HTTP API and provides 18 read-only tools covering instant and range PromQL evaluations, metric and label discovery, series tracking, target status, TSDB statistics, alerting rules, and exemplars. When deployed within a Kubernetes cluster or standard infrastructure, MCP Prometheus can operate as an OAuth 2.1 authorization server integrated with upstream providers like Dex or Google. This setup enforces tenant-isolated metrics access by resolving tenant identifiers for authenticated users before queries reach Mimir or Prometheus. AI agents like Claude can run diagnostics, check active alerts, and inspect monitoring targets across production systems securely while respecting existing organization permissions and avoiding uncontrolled access to sensitive time-series data.

Category: Data & Analytics

Tags: metrics, monitoring, prometheus, time-series

Visit MCP Prometheus

How to install and configure MCP Prometheus

1. Build the binary from source using Go: bash git clone https://github.com/giantswarm/mcp-prometheus.git cd mcp-prometheus go build -o mcp-prometheus ./... Alternatively, download a pre-built binary from the GitHub repository release page. 2. Set the connection environment variables, such as PROMETHEUS_URL pointing to your Prometheus or Mimir instance, along with PROMETHEUS_TOKEN or PROMETHEUS_USERNAME and PROMETHEUS_PASSWORD if authentication is enabled. 3. For desktop clients such as Claude Desktop, configure the client configuration file to start the server over stdio transport: json { "mcpServers": { "prometheus": { "command": "/path/to/mcp-prometheus", "args": ["serve", "--transport", "stdio"], "env": { "PROMETHEUS_URL": "http://localhost:9090" } } } } 4. For cluster installations using HTTP and OAuth, launch the process with ./mcp-prometheus serve --transport streamable-http --http-addr :8080 --enable-oauth and register the resulting HTTP URL with your MCP client.

What you can do with MCP Prometheus

• Executing instant PromQL queries to check real-time CPU, memory, and disk usage across Kubernetes clusters directly from Claude. - Running PromQL range queries over specific timeframes to analyze historical traffic spikes and latency trends during incident investigations. - Discovering available metric names, labels, and series to help engineers identify monitoring coverage without searching Grafana dashboards manually. - Inspecting active firing alerts and rule evaluation states across Prometheus installations to triage ongoing production operational incidents quickly. - Reviewing scrape targets and TSDB runtime statistics to diagnose unhealthy endpoints and monitor the health of Prometheus itself.

Key facts

• Open Source
• https://github.com/giantswarm/mcp-prometheus
• Data & Analytics, Monitoring & Observability
• metrics, monitoring, prometheus, time-series

Part of MCP Servers

What can MCP Prometheus do?

MCP Prometheus allows AI assistants to interact directly with Prometheus and Mimir monitoring environments. It offers eighteen read-only tools that let models evaluate instant and range PromQL queries, discover metric names and label values, inspect active alerting rules, review scrape target health, examine TSDB statistics, and extract exemplars. This lets models answer operational questions directly from live telemetry.

Which MCP clients work with MCP Prometheus?

MCP Prometheus functions with any client implementing the Model Context Protocol. For local workstations, desktop applications such as Claude Desktop or MCP Inspector can execute it via standard input and output transport. For remote cluster deployments, clients supporting SSE or streamable HTTP transports can connect directly while authenticating over OAuth 2.1 tokens.

How does MCP Prometheus handle multi-tenancy?

MCP Prometheus provides multi-tenant access controls for Mimir environments. When configured with Dex or Google identity providers, the server verifies incoming OAuth tokens and resolves the user identity against Grafana organizations or static group mappings. It then attaches the appropriate tenant headers to outgoing queries, ensuring users only inspect metrics belonging to their authorized organizations.

Is MCP Prometheus open source?

Yes, MCP Prometheus is open source software developed and maintained by Giant Swarm. The complete source code, Helm charts, documentation, and pre-built binaries are distributed freely under an open-source license on GitHub at https://github.com/giantswarm/mcp-prometheus. Anyone can inspect, build, or contribute to the project.