Prometheus MCP Server

What pab1it0/prometheus-mcp-server MCP server does

The pab1it0/prometheus-mcp-server MCP server gives MCP-compatible AI clients access to a Prometheus instance. It is intended for querying monitoring data, inspecting the metrics available in a Prometheus installation, and checking scrape-target status without exposing Prometheus operations as custom client-specific integrations.

The server provides six documented tools. execute_query runs an instant PromQL query, while execute_range_query evaluates PromQL over a specified start time, end time, and step interval. list_metrics supports paginated and filtered metric discovery. get_metric_metadata retrieves metadata for one metric or multiple metrics, with optional filtering. get_targets returns scrape targets and supports filtering by target state or scrape pool, along with optional pagination. health_check supports container monitoring and status verification.

How it works

At runtime, the server connects to the Prometheus URL supplied through configuration and presents the available operations to an MCP client. Query results can include Prometheus UI links unless link generation is disabled. The request timeout defaults to 30 seconds, and the server can be configured to use standard input/output, HTTP, or SSE transport.

The tool set is configurable. Operators can omit tools that are not needed, reducing the functionality presented to the client and potentially limiting the context used by an AI assistant. TOOL_PREFIX can add a prefix to every tool name, which is useful when multiple instances target different environments.

Setup and configuration

The pab1it0/prometheus-mcp-server MCP server requires a reachable Prometheus installation and an MCP-compatible client. The documented container image is ghcr.io/pab1it0/prometheus-mcp-server:latest, and the project also publishes a Helm chart through an OCI registry for Kubernetes deployment.

PROMETHEUS_URL is required. Optional connection settings include SSL verification control, request timeout, custom headers, and REQUESTS_CA_BUNDLE for TLS certificate verification. Prometheus authentication can use PROMETHEUS_USERNAME and PROMETHEUS_PASSWORD for basic authentication, PROMETHEUS_TOKEN for bearer authentication, or client certificate and key paths for mutual TLS. ORG_ID is available for multi-tenant configurations.

For non-stdio operation, PROMETHEUS_MCP_SERVER_TRANSPORT selects stdio, http, or sse. HTTP binding can be adjusted with PROMETHEUS_MCP_BIND_HOST and PROMETHEUS_MCP_BIND_PORT; stateless HTTP mode is available for multi-replica deployments.

Tools and capabilities

• Run instant PromQL queries against Prometheus.
• Run time-range PromQL queries with caller-supplied intervals and step sizes.
• Enumerate metrics with pagination and filtering.
• Retrieve individual or bulk metric metadata.
• Search metadata with optional metric-related filters.
• Inspect scrape targets with state and scrape-pool filters.
• Check server health for container status monitoring.

Limitations and notes

The pab1it0/prometheus-mcp-server MCP server requires access to an existing Prometheus server; it does not provide a Prometheus storage or monitoring deployment itself. The material documents container and Helm distribution, but does not specify a native package-manager command for installing the MCP server directly into a local Python environment.

Authentication is configuration-dependent rather than fixed to one credential type. Basic authentication, bearer tokens, mutual TLS, custom headers, and organization IDs are supported as optional settings. The default MCP transport is stdio, while HTTP and SSE require the corresponding transport configuration.