msrashed-prometheus-mcp-server 0.1.0

Read-only Prometheus MCP server for metrics querying, alerting, and monitoring inspection

Prometheus MCP Server

A read-only Model Context Protocol (MCP) server for Prometheus, enabling AI agents like Claude to query metrics, investigate alerts, and analyze monitoring data safely.

Features

• Read-Only Safety: All operations are read-only. No modifications to Prometheus configuration or data
• Comprehensive Query Support: Execute PromQL instant and range queries
• Metadata Discovery: List metrics, labels, and series
• Target Monitoring: View scrape targets and their health
• Alert Investigation: List active alerts and alert rules
• Configuration Inspection: View Prometheus configuration and runtime info
• Multiple Auth Methods: Support for Bearer tokens and Basic authentication
• Type-Safe: Full type hints for Python 3.12+

Installation

Using uv (Recommended)

# Clone the repository
git clone <repository-url>
cd prometheus-mcp-server

# Install with uv
uv pip install -e .

# Or install from PyPI (when published)
uv pip install prometheus-mcp-server

Using pip

pip install prometheus-mcp-server

Using pipx (for CLI usage)

pipx install prometheus-mcp-server

Quick Start

1. Set Environment Variables

# Prometheus server URL (required)
export PROM_URL="http://localhost:9090"

# Optional: Bearer token authentication
export PROM_TOKEN="your_bearer_token"

# Optional: Basic authentication
export PROM_USERNAME="admin"
export PROM_PASSWORD="secret"

# Optional: Timeout and SSL settings
export PROM_TIMEOUT="30"
export PROM_VERIFY_SSL="true"

2. Run the Server

# Using stdio transport (default, for Claude Desktop)
prometheus-mcp-server

# Using HTTP transport
prometheus-mcp-server --transport http --port 8000

# With custom Prometheus URL
prometheus-mcp-server --url https://prometheus.example.com

3. Configure with Claude Desktop

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "prometheus": {
      "command": "prometheus-mcp-server",
      "env": {
        "PROM_URL": "http://localhost:9090",
        "PROM_TOKEN": "optional_bearer_token"
      }
    }
  }
}

Or with uvx:

{
  "mcpServers": {
    "prometheus": {
      "command": "uvx",
      "args": ["prometheus-mcp-server"],
      "env": {
        "PROM_URL": "http://localhost:9090"
      }
    }
  }
}

Available Tools

Query Tools

query_instant

Execute a PromQL instant query at a single point in time.

# Check which targets are up
query_instant(query="up")

# Get API server status
query_instant(query='up{job="api-server"}', time="now")

# Get current request rate
query_instant(query="rate(http_requests_total[5m])")

query_range

Execute a PromQL query over a time range.

# Get request rate over last hour
query_range(
    query="rate(http_requests_total[5m])",
    start="now-1h",
    end="now",
    step="30s"
)

# Get CPU usage by instance
query_range(
    query="avg(cpu_usage) by (instance)",
    start="2024-01-15T00:00:00Z",
    end="2024-01-15T12:00:00Z",
    step="1m"
)

query_exemplars

Query exemplars for trace correlation.

query_exemplars(
    query="http_request_duration_seconds_bucket",
    start="now-1h",
    end="now"
)

Metadata Discovery Tools

list_metrics

List all available metrics.

# List all metrics
list_metrics()

# List HTTP-related metrics
list_metrics(match="http_.*")

# List all counter metrics
list_metrics(match=".*_total")

get_metric_metadata

Get metadata (type, help text) for metrics.

# Get metadata for a specific metric
get_metric_metadata(metric="http_requests_total")

# Get metadata for all metrics (limited to 100)
get_metric_metadata(limit=100)

list_labels

List all label names.

# List all labels
list_labels()

# List labels for specific job
list_labels(match=['{job="api"}'])

get_label_values

Get all values for a specific label.

# Get all job names
get_label_values(label="job")

# Get namespaces in prod cluster
get_label_values(
    label="namespace",
    match=['{cluster="prod"}']
)

find_series

Find time series matching label selectors.

# Find all series for a job
find_series(match=['{job="api"}'])

# Find all HTTP metrics in production
find_series(
    match=['{__name__=~"http_.*",env="production"}'],
    start="now-1h",
    end="now"
)

Target & Scrape Tools

list_targets

List all scrape targets and their status.

# List all targets
list_targets()

# List only active targets
list_targets(state="active")

# List only dropped targets
list_targets(state="dropped")

get_targets_metadata

Get metadata about metrics from targets.

# Get metadata for specific target
get_targets_metadata(match_target='{job="api-server"}')

# Get metadata for specific metric
get_targets_metadata(metric="http_requests_total")

Alert Tools

list_alerts

List all active alerts (firing and pending).

list_alerts()

list_rules

List all recording and alerting rules.

# List all rules
list_rules()

# List only alerting rules
list_rules(type="alert")

# List only recording rules
list_rules(type="record")

Configuration & Status Tools

get_config

Get the current Prometheus configuration.

get_config()

get_flags

Get Prometheus runtime flags.

get_flags()

get_runtime_info

Get Prometheus runtime information.

get_runtime_info()

get_tsdb_stats

Get TSDB statistics and cardinality.

get_tsdb_stats()

check_health

Check Prometheus health status.

check_health()

check_readiness

Check if Prometheus is ready to serve queries.

check_readiness()

PromQL Query Examples

Basic Queries

# Check target health
up

# Filter by job
up{job="api-server"}

# Get metric value
http_requests_total

# Multiple label filters
http_requests_total{job="api",status="200"}

Rate Queries

# Request rate over 5 minutes
rate(http_requests_total[5m])

# Sum rate by status code
sum(rate(http_requests_total[5m])) by (status)

# Instant rate (more sensitive to spikes)
irate(cpu_seconds_total[1m])

Aggregation

# Average CPU by instance
avg(cpu_usage) by (instance)

# Total memory by namespace
sum(memory_usage) by (namespace)

# Maximum response time by endpoint
max(response_time) by (endpoint)

# Count of targets by job
count(up) by (job)

Advanced Queries

# 95th percentile response time
histogram_quantile(0.95, rate(http_request_duration_seconds_bucket[5m]))

# Predict disk usage in 1 hour
predict_linear(disk_usage[1h], 3600)

# Temperature change over 5 minutes
delta(cpu_temp[5m])

# Detect rate of increase
deriv(cpu_temp[5m])

Time Range Queries

# CPU usage increase over last hour vs previous hour
(avg_over_time(cpu_usage[1h]) - avg_over_time(cpu_usage[1h] offset 1h))

# Compare to last week
http_requests_total - http_requests_total offset 1w

Authentication

Bearer Token Authentication

export PROM_URL="https://prometheus.example.com"
export PROM_TOKEN="your_bearer_token"
prometheus-mcp-server

Basic Authentication

export PROM_URL="https://prometheus.example.com"
export PROM_USERNAME="admin"
export PROM_PASSWORD="secret"
prometheus-mcp-server

Kubernetes Service Account Token

# Get token from Kubernetes
TOKEN=$(kubectl get secret -n monitoring prometheus-token -o jsonpath='{.data.token}' | base64 -d)

export PROM_URL="https://prometheus.monitoring.svc.cluster.local:9090"
export PROM_TOKEN="$TOKEN"
prometheus-mcp-server

Configuration

Environment Variables

Variable	Description	Default	Required
PROM_URL	Prometheus server URL	http://localhost:9090	No
PROMETHEUS_URL	Alternative to PROM_URL	-	No
PROM_TOKEN	Bearer token for auth	-	No
PROMETHEUS_TOKEN	Alternative to PROM_TOKEN	-	No
PROM_USERNAME	Username for basic auth	-	No
PROM_PASSWORD	Password for basic auth	-	No
PROM_TIMEOUT	Request timeout (seconds)	30	No
PROM_VERIFY_SSL	Verify SSL certificates	true	No

Command-Line Arguments

prometheus-mcp-server \
  --url https://prometheus.example.com \
  --token your_bearer_token \
  --timeout 60 \
  --transport http \
  --port 8000

Full options:

--url URL                  Prometheus server URL
--token TOKEN              Bearer token for authentication
--username USERNAME        Username for basic auth
--password PASSWORD        Password for basic auth
--timeout SECONDS          Request timeout in seconds (default: 30)
--no-verify-ssl            Disable SSL verification (not recommended)
--transport {stdio,http,sse}  Transport mechanism (default: stdio)
--host HOST                Host for HTTP/SSE transport (default: 127.0.0.1)
--port PORT                Port for HTTP/SSE transport (default: 8000)

Use Cases

Incident Investigation

Ask questions like:

• "What's the current CPU usage across all pods?"
• "Show me the error rate for the API service in the last hour"
• "Which targets are down right now?"
• "What alerts are currently firing?"

Performance Analysis

• "Compare request latency between now and 24 hours ago"
• "Show me the top 10 endpoints by request volume"
• "What's the memory usage trend for the worker pods?"

Capacity Planning

• "What's the 95th percentile response time over the last week?"
• "Show me the disk usage growth rate"
• "Which services have the highest cardinality?"

Alert Analysis

• "Why is the HighMemoryUsage alert firing?"
• "Show me the history of the DiskSpaceLow alert"
• "What's the current state of all alerting rules?"

Architecture

Project Structure

prometheus-mcp-server/
├── src/
│   └── prometheus_mcp_server/
│       ├── __init__.py          # Package initialization
│       ├── __main__.py          # CLI entry point
│       ├── server.py            # FastMCP server setup
│       ├── tools/
│       │   ├── __init__.py
│       │   └── registry.py      # All tool implementations
│       └── utils/
│           ├── __init__.py
│           ├── client.py        # Prometheus HTTP client
│           └── helpers.py       # Time parsing, formatting
├── pyproject.toml               # Project configuration
├── README.md                    # This file
└── USER_STORIES.md              # Detailed requirements

Technology Stack

• MCP Framework: FastMCP 2.0
• HTTP Client: httpx (async support)
• Python: 3.12+ with full type hints
• Authentication: Bearer token and Basic auth support
• Read-Only: No write operations allowed

Development

Setup Development Environment

# Clone repository
git clone <repository-url>
cd prometheus-mcp-server

# Install with dev dependencies using uv
uv pip install -e ".[dev]"

# Or with pip
pip install -e ".[dev]"

Run Tests

# Run all tests
pytest

# Run with coverage
pytest --cov=prometheus_mcp_server

# Run specific test file
pytest tests/test_client.py

Code Quality

# Format code
ruff format .

# Lint code
ruff check .

# Fix linting issues
ruff check --fix .

# Type checking (if using mypy)
mypy src/

Testing with Local Prometheus

# Run Prometheus locally with Docker
docker run -d -p 9090:9090 prom/prometheus

# Test the MCP server
export PROM_URL="http://localhost:9090"
prometheus-mcp-server

Security

Read-Only Guarantee

This MCP server is designed to be strictly read-only:

• Only GET requests and specific read-only POST endpoints are allowed
• All write operations (PUT, DELETE, PATCH) are blocked
• No configuration modifications possible
• No data deletion or manipulation

Blocked Operations

The following operations are blocked by design:

• Creating or deleting targets
• Modifying alert rules
• Changing Prometheus configuration
• Deleting time series data
• Administrative operations

Safe Endpoints Only

Only these endpoint patterns are allowed:

• GET /api/v1/* - All read operations
• POST /api/v1/query* - Query operations only
• POST /api/v1/series - Series discovery only
• POST /api/v1/labels - Label queries only
• GET /-/healthy - Health checks
• GET /-/ready - Readiness checks

Troubleshooting

Connection Issues

# Test Prometheus connectivity
curl -s http://localhost:9090/api/v1/query?query=up

# Check with authentication
curl -H "Authorization: Bearer $PROM_TOKEN" \
  https://prometheus.example.com/api/v1/query?query=up

SSL Certificate Issues

# Disable SSL verification (not recommended for production)
export PROM_VERIFY_SSL="false"
prometheus-mcp-server

# Or use command-line flag
prometheus-mcp-server --no-verify-ssl

Timeout Issues

# Increase timeout for slow queries
export PROM_TIMEOUT="60"
prometheus-mcp-server

# Or use command-line flag
prometheus-mcp-server --timeout 60

Debug Mode

Enable debug logging:

# Set log level
export LOG_LEVEL="DEBUG"
prometheus-mcp-server

Examples

Example Configuration Files

Claude Desktop Config (macOS)

~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "prometheus-local": {
      "command": "prometheus-mcp-server",
      "env": {
        "PROM_URL": "http://localhost:9090"
      }
    },
    "prometheus-prod": {
      "command": "prometheus-mcp-server",
      "env": {
        "PROM_URL": "https://prometheus.prod.example.com",
        "PROM_TOKEN": "prod_bearer_token",
        "PROM_TIMEOUT": "60"
      }
    }
  }
}

Linux Config

~/.config/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "prometheus": {
      "command": "/home/user/.local/bin/prometheus-mcp-server",
      "env": {
        "PROM_URL": "http://localhost:9090"
      }
    }
  }
}

Example Queries

Investigate High Memory Alert

User: "The HighMemoryUsage alert is firing. Help me investigate."

AI uses:
1. list_alerts() - See all active alerts
2. query_instant(query='container_memory_usage{job="api"}') - Check current usage
3. query_range(query='container_memory_usage{job="api"}', start="now-6h", end="now") - See trend
4. list_rules(type="alert") - Check alert threshold

Find Top CPU Consumers

User: "Which pods are using the most CPU?"

AI uses:
1. query_instant(query='topk(10, rate(container_cpu_usage_seconds_total[5m]))') - Top 10 CPU users
2. get_label_values(label="pod") - List all pods
3. query_range(...) - Check trend over time

Check Service Health

User: "Is the API service healthy?"

AI uses:
1. query_instant(query='up{job="api-server"}') - Check if targets are up
2. list_targets(state="active") - See all active targets
3. query_instant(query='rate(http_requests_total{job="api",status=~"5.."}[5m])') - Check error rate
4. list_alerts() - Check for any alerts

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Ensure code passes ruff checks
5. Submit a pull request

License

MIT License - see LICENSE file for details

Support

• Issues: GitHub Issues
• Prometheus Docs: https://prometheus.io/docs/
• PromQL Guide: https://prometheus.io/docs/prometheus/latest/querying/basics/
• MCP Documentation: https://modelcontextprotocol.io/

Related Projects

• Datadog MCP Server - MCP server for Datadog
• Grafana MCP Server - MCP server for Grafana

Acknowledgments

• Built with FastMCP
• Follows Model Context Protocol specification
• Inspired by the Prometheus community's excellent API design