Datadog Model Context Protocol (MCP) 🔍

A Python-based tool to interact with Datadog API and fetch monitoring data from your infrastructure. This MCP provides easy access to monitor states and Kubernetes logs through a simple interface.

Datadog Features 🌟

Monitor State Tracking: Fetch and analyze specific monitor states
Kubernetes Log Analysis: Extract and format error logs from Kubernetes clusters

Prerequisites 📋

Python 3.11+
Datadog API and Application keys (with correct permissions)
Access to Datadog site

Installation 🔧

pip install -r requirements.txt

Required packages:

datadog-api-client
fastmcp
loguru
icecream
python-dotenv
uv

Environment Setup 🔑

Create a .env file with your Datadog credentials:

DD_API_KEY=your_api_key
DD_APP_KEY=your_app_key

Setup Claude Desktop Setup for MCP 🖥️

Install Claude Desktop

# Assuming you're on macOS
brew install claude-desktop

# Or download from official website
https://claude.ai/desktop

Set up Datadog MCP config:

# on mac is 
~/Library/Application\ Support/Claude/claude_desktop_config.json


# Add this to your claude config json
```json
    "Datadog-MCP-Server": {
      "command": "uv",
      "args": [
        "run",
        "--with",
        "datadog-api-client",
        "--with",
        "fastmcp",
        "--with",
        "icecream",
        "--with",
        "loguru",
        "--with",
        "python-dotenv",
        "fastmcp",
        "run",
        "/your-path/mcp-collection/datadog/main.py"
      ],
      "env": {
        "DD_API_KEY": "xxxx",
        "DD_APP_KEY": "xxx"
      }
    },

Usage 💻

Architecture 🏗

FastMCP Base: Utilizes FastMCP framework for tool management
Modular Design: Separate functions for monitors and logs
Type Safety: Full typing support with Python type hints
API Abstraction: Wrapped Datadog API calls with error handling

I'll add a section about MCP and Claude Desktop setup:

Model Context Protocol (MCP) Introduction 🤖

What is MCP?

Model Context Protocol (MCP) is a framework allowing AI models to interact with external tools and APIs in a standardized way. It enables models like Claude to:

Access external data
Execute commands
Interact with APIs
Maintain context across conversations

some examples of MCP servers

https://github.com/punkpeye/awesome-mcp-servers?tab=readme-ov-file

Tutorial for setup MCP

https://medium.com/@pedro.aquino.se/how-to-use-mcp-tools-on-claude-desktop-app-and-automate-your-daily-tasks-1c38e22bc4b0

How it works - Available Functions 🛠️

the LLM use provided function to get the data and use it

1. Get Monitor States

get_monitor_states(
    name: str,           # Monitor name to search
    timeframe: int = 1   # Hours to look back
)

Example:

response = get_monitor_states(name="traefik")

# Sample Output
{
    "id": "12345678",
    "name": "traefik",
    "status": "OK",
    "query": "avg(last_5m):avg:traefik.response_time{*} > 1000",
    "message": "Response time is too high",
    "type": "metric alert",
    "created": "2024-01-14T10:00:00Z",
    "modified": "2024-01-14T15:30:00Z"
}

2. Get Kubernetes Logs

get_k8s_logs(
    cluster: str,            # Kubernetes cluster name
    timeframe: int = 5,      # Hours to look back
    namespace: str = None    # Optional namespace filter
)

Example:

logs = get_k8s_logs(
    cluster="prod-cluster",
    timeframe=3,
    namespace="default"
)

# Sample Output
{
    "timestamp": "2024-01-14T22:00:00Z",
    "host": "worker-1",
    "service": "nginx-ingress",
    "pod_name": "nginx-ingress-controller-abc123",
    "namespace": "default",
    "container_name": "controller",
    "message": "Connection refused",
    "status": "error"
}

# Install as MCP extension
cd datadog
task install-mcp

4. Verify Installation

In Claude chat desktop

check datadog connection in claude

5. Use Datadog MCP Tools

Security Considerations 🔒

Store API keys in .env
MCP runs in isolated environment
Each tool has defined permissions
Rate limiting is implemented

Troubleshooting 🔧

Using MCP Inspector

# Launch MCP Inspector for debugging
task run-mcp-inspector

The MCP Inspector provides:

Real-time view of MCP server status
Function call logs
Error tracing
API response monitoring

Common issues and solutions

API Authentication Errors
Error: (403) Forbidden
➡️ Check your DD_API_KEY and DD_APP_KEY in .env
MCP Connection Issues
Error: Failed to connect to MCP server
➡️ Verify your claude_desktop_config.json path and content
Monitor Not Found
Error: No monitor found with name 'xxx'
➡️ Check monitor name spelling and case sensitivity
logs can be found here

Contributing 🤝

Feel free to:

Open issues for bugs
Submit PRs for improvements
Add new features

Notes 📝

API calls are made to Datadog EU site
Default timeframe is 1 hour for monitor states
Page size limits are set to handle most use cases