home / mcp / prometheus mcp server

Prometheus MCP Server

Provides an MCP gateway to Prometheus metrics collected via OpenTelemetry, with an HTTP interface for MCP clients to query metrics and generate PromQL.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "fabriciodf-otel_prom_mcp_server": {
      "url": "http://localhost:8082",
      "headers": {
        "LLAMA_MODEL": "llama3.2:1b",
        "PROMETHEUS_URL": "http://prometheus:9090",
        "PROMETHEUS_MCP_BIND_HOST": "0.0.0.0",
        "PROMETHEUS_MCP_BIND_PORT": "8082",
        "PROMETHEUS_MCP_SERVER_TRANSPORT": "http"
      }
    }
  }
}

This MCP server stack collects metrics from a FastAPI-based demo, enables PromQL generation via an LLM-backed UI, and exposes an MCP-compatible HTTP interface to query the Prometheus instance. It ties together OpenTelemetry metrics, a Prometheus-backed data plane, and an MCP client workflow for programmatic access to metrics data and PromQL insights.

How to use

You interact with the Prometheus MCP Server by connecting an MCP client to the HTTP endpoint exposed by the stack. Your client can query the Prometheus data available via the MCP server, or request PromQL expressions generated from natural language prompts through the UI. Typical usage flows include generating PromQL from natural language queries, running those queries against Prometheus through the MCP gateway, and inspecting the resulting metrics data.

How to install

Prerequisites you need before starting: Docker and Docker Compose installed on your machine. You should also have sufficient disk space to fetch the Ollama model (approximately 1–2 GB for llama3.2:1b).

Step by step commands you will run to set up and start the stack:

cd prometheus-ai
./scripts/install.sh
# or to skip model download if you already have it
./scripts/install.sh --skip-model-pull

# or perform a full reset (pulls images, rebuilds, starts, and pulls the Ollama model)
./run.sh

# If you did not use the automatic run script, start the services with Docker Compose
docker compose up -d

# If you skipped the Ollama model initially, pull it afterwards
docker exec prometheus-ai-ollama ollama pull llama3.2:1b

Additional notes

Endpoint overview for quick access after the stack is running:

Demo API: http://localhost:8000/docs (health and item endpoints)

Prompt UI: http://localhost:8080 (generates PromQL via Llama3 and shows JSON and a natural-language interpretation in PT-BR)

Prometheus: http://localhost:9090

OTel Collector endpoints: OTLP gRPC http://localhost:4317, HTTP http://localhost:4318, metrics http://localhost:9464/metrics

MCP Server (HTTP): http://localhost:8082

Configuration and environment

This stack relies on environment configuration to wire the Prometheus instance into the MCP gateway and to control the Ollama-based Llama3 model. The primary MCP-facing configuration is exposed as environment variables when starting the MCP server container.

PROMETHEUS_URL=http://prometheus:9090
PROMETHEUS_MCP_SERVER_TRANSPORT=http
PROMETHEUS_MCP_BIND_HOST=0.0.0.0
PROMETHEUS_MCP_BIND_PORT=8082

MCP server configuration (HTTP)

The MCP server is exposed over HTTP and connects to the Prometheus instance to serve metrics data and generated PromQL to MCP clients.

{
  "mcpServers": {
    "prometheus": {
      "type": "http",
      "name": "prometheus_mcp",
      "url": "http://localhost:8082",
      "args": []
    }
  }
}

Observability and troubleshooting

Common metrics you may see include the app-level metrics like demo_requests_total, demo_request_latency_ms, and demo_pending_orders, as well as standard HTTP/runtime metrics such as process_cpu_seconds_total and up. If you encounter issues, check live logs for app-service, ui-service, and otel-collector to trace data flow from the Demo API through the OpenTelemetry collector to Prometheus.

docker compose logs -f app-service ui-service otel-collector

APIs and endpoints

The stack exposes a set of endpoints for metrics and UI interactions. The MCP server endpoint is the primary entry for programmatic access via MCP clients.

Available tools

demo_api_metrics

Expose and collect application metrics through OpenTelemetry in the Demo API, including requests, latency, and order processing metrics.

prompt_ui_promql

UI component that uses an LLM (Ollama/Llama3) to generate PromQL from natural language prompts and display results.

ollama_llama3

Ollama-based Llama3 model used to translate natural language questions into PromQL queries.

prometheus_mcp_server

MCP gateway that exposes an HTTP interface to query Prometheus data and PromQL expressions via MCP clients.