Prometheus MCP Server

A Model Context Protocol server that enables AI agents to interact directly with Prometheus metrics data through natural language queries.

Features

This MCP server provides direct integration with any Prometheus instance, allowing AI assistants to:

• Query Metrics: Execute PromQL queries for instant and range data

• Explore Metadata: Discover available metrics, labels, and their values

• Analyze Time Series: Get detailed information about specific metric series

• Natural Language Interface: Ask questions about your metrics in plain English

Related MCP server: Prometheus MCP Server

Available Tools

prometheus_query

Execute instant PromQL queries at a specific point in time.

Example questions:

• "What's the current CPU usage?"

• "Show me the memory consumption right now"

• "How many requests per second is my service handling?"

prometheus_query_range

Execute range queries to get time series data over a period.

Example questions:

• "Show me CPU usage over the last hour"

• "What was the request rate trend for the past 24 hours?"

• "Graph the error rate for my API over the last week"

prometheus_labels

Get the list of available labels for metrics.

Example questions:

• "What labels are available for the http_requests_total metric?"

• "Show me all labels for metrics matching 'cpu'"

prometheus_label_values

Get all possible values for a specific label.

Example questions:

• "What are all the possible values for the 'environment' label?"

• "List all services that have metrics"

• "What instances are reporting metrics?"

prometheus_series

Get metadata about time series matching specific selectors.

Example questions:

• "Show me all time series for the user service"

• "What metrics exist for my database?"

prometheus_metadata

Get help text and type information about metrics.

Example questions:

• "What does the node_cpu_seconds_total metric measure?"

• "Show me documentation for all available metrics"

Installation

Using NPM (Recommended)

# Install globally
npm install -g @prometheus-mcp/server

# Or run directly with npx
npx @prometheus-mcp/server

Building from Source

# Clone the repository
git clone https://github.com/ramkrsna/prometheus-mcp-server.git
cd prometheus-mcp-server

# Build the binary
go build -o prometheus-mcp-server

# Or install directly
go install

Testing with Mock Data (Podman)

# Install Podman (if not already installed)
brew install podman podman-compose

# Start mock environment
./start-mock-env.sh

# Configure and run MCP server
export PROMETHEUS_URL=http://localhost:9090
./prometheus-mcp-server

Configuration

The server requires the following environment variables:

• PROMETHEUS_URL (required): URL of your Prometheus instance (e.g., http://localhost:9090)

• PROMETHEUS_USERNAME (optional): Username for basic authentication

• PROMETHEUS_PASSWORD (optional): Password for basic authentication

• HTTP_MODE (optional): Set to true to run in HTTP/WebSocket mode instead of STDIO

• HOST (optional): HTTP server host (default: 0.0.0.0)

• PORT (optional): HTTP server port (default: 8080)

Usage with AI Assistants

Claude Desktop

1. Open Claude Desktop settings

2. Go to Developer → Edit Config

3. Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "prometheus": {
      "command": "prometheus-mcp",
      "env": {
        "PROMETHEUS_URL": "http://localhost:9090",
        "PROMETHEUS_USERNAME": "admin",
        "PROMETHEUS_PASSWORD": "password"
      }
    }
  }
}

Cursor

1. Open Cursor settings

2. Go to Features → MCP

3. Add the server configuration:

{
  "mcpServers": {
    "prometheus": {
      "command": "prometheus-mcp",
      "env": {
        "PROMETHEUS_URL": "http://localhost:9090"
      }
    }
  }
}

VS Code

1. Open VS Code settings

2. Search for "MCP"

3. Add to your settings.json:

{
  "mcp.servers": {
    "prometheus": {
      "type": "stdio",
      "command": "prometheus-mcp",
      "env": {
        "PROMETHEUS_URL": "http://localhost:9090"
      }
    }
  }
}

Example Conversations

Basic Metrics Query

You: "What's the current memory usage across all my nodes?"

AI: I'll query the memory usage for all nodes. Uses prometheus_query with: (1 - (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes)) * 100

Here's the current memory usage:

• node1: 45.2%

• node2: 62.8%

• node3: 38.5%

Troubleshooting High Load

You: "My service seems slow, can you help me investigate?"

AI: I'll help you investigate the performance issue. Let me check a few things:

Uses prometheus_query_range for response time Uses prometheus_query for current request rate Uses prometheus_query for error rate

I found that:

1. Your response time increased from 200ms to 850ms in the last hour

2. Request rate is 3x higher than usual (15k req/s vs normal 5k req/s)

3. Error rate is still low at 0.1%

This appears to be a load-related issue rather than errors.

Exploring Metrics

You: "What metrics do I have available for my PostgreSQL database?"

AI: Let me explore the PostgreSQL metrics in your Prometheus instance.

Uses prometheus_series with match: {job="postgresql"}

You have the following PostgreSQL metrics available:

• pg_stat_database_tup_inserted - Number of rows inserted

• pg_stat_database_tup_updated - Number of rows updated

• pg_stat_database_conflicts - Number of conflicts

• pg_stat_activity_count - Number of active connections

• pg_database_size_bytes - Database size in bytes [... and 23 more metrics]

Development

Project Structure

prometheus-mcp-server/
├── main.go                    # Entry point
├── tools.go                   # MCP tool definitions
├── http_server.go            # HTTP/WebSocket transport
├── internal/
│   ├── models/               # Configuration models
│   ├── prometheus/           # Prometheus client and handlers
│   └── utils/                # Utilities
├── bin/                      # NPM distribution scripts
└── package.json              # NPM package metadata

Adding New Tools

1. Add the tool definition in tools.go

2. Create the handler in internal/prometheus/handlers.go

3. Add any new Prometheus API calls in internal/prometheus/client.go

Testing

# Run tests
go test ./...

# Test with a local Prometheus
PROMETHEUS_URL=http://localhost:9090 go run .

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and feature requests, please use the GitHub issue tracker.