MCP Datadog Server

A Model Context Protocol (MCP) server that exposes Datadog APIs to AI assistants and code editors via tools.

Overview

The server provides MCP tools for:

• Metrics – Query metrics, metadata, list metrics

• Logs – Search logs, get log details, aggregate logs

• Events – Search events, get event details

• Monitors – List monitors, get status, search monitors

• APM/Traces – Query traces, service health, service dependencies

• Services – Service dependencies (single and multi-environment)

Related MCP server: MCP Datadog Server

Quick Start

Prerequisites

• Node.js 22+

• npm

Run with npx (recommended)

No clone or install needed. Add the server to your MCP client (e.g. Cursor, Claude) using stdio and run it from GitHub:

{
  "datadog": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "github:micaelmalta/mcp-server-datadog"],
    "env": {
      "DATADOG_API_KEY": "your_api_key",
      "DATADOG_APP_KEY": "your_app_key"
    }
  }
}

Set DATADOG_API_KEY and DATADOG_APP_KEY (and optionally DATADOG_SITE, default datadoghq.com). Restart the client so the tools appear.

To try from the terminal:

DATADOG_API_KEY=your_key DATADOG_APP_KEY=your_app_key npx -y github:micaelmalta/mcp-server-datadog

Run from source

For development or a fixed install:

git clone https://github.com/micaelmalta/mcp-server-datadog.git
cd mcp-server-datadog
npm install
cp .env.example .env

Edit .env and set DATADOG_API_KEY and DATADOG_APP_KEY. Then:

npm start
# or with NODE_ENV=local: npm run dev

In your MCP config, use stdio with node and the path to the entry point:

{
  "datadog": {
    "type": "stdio",
    "command": "node",
    "args": ["/path/to/mcp-server-datadog/src/index.js"],
    "env": {
      "DATADOG_API_KEY": "your_api_key",
      "DATADOG_APP_KEY": "your_app_key"
    }
  }
}

Tools

Example prompts: "Show error logs from service X in the last hour" → search_logs. "What's CPU usage on production?" → query_metrics. "How is the API service doing?" → get_service_health.

Time ranges: Use ISO 8601 or Unix timestamps (seconds for metrics/events, milliseconds for logs/APM). Filters: Datadog syntax, e.g. service:api, status:error, env:production.

Project structure

mcp_datadog/
├── src/
│   ├── clients/     # Datadog API clients (SDK-based)
│   ├── tools/       # MCP tool definitions and handlers
│   ├── utils/       # Environment, errors, logger, toolErrors
│   └── index.js     # Server entry point
├── test/            # Vitest tests and fixtures
│   ├── benchmark/   # Tool handler benchmarks (mocked)
│   ├── mocks/      # Datadog SDK mocks
│   └── ...
├── docs/            # Additional documentation
└── package.json

Tech stack: Node.js 22+, JavaScript (ESM), JSDoc, Vitest, ESLint, Prettier, @modelcontextprotocol/sdk, @datadog/datadog-api-client.

Development

Commands

API client pattern

Clients return { data, error }:

const { data, error } = await client.queryMetrics(query, from, to);
if (error) {
  console.error(error.message, error.statusCode);
} else {
  console.log(data);
}

CI

GitHub Actions (.github/workflows/ci.yml) runs on push and pull requests to main/master: format check (prettier --check), lint (ESLint), and tests. Same commands locally: npm run format:check && npm run lint && npm test.

Testing

Tests use Vitest with mocked Datadog SDK (test/mocks/datadogApi.js) and fixtures in test/fixtures/. Run npm test before committing.

E2E tests (test/e2e/) run against the real Datadog API. They are skipped unless RUN_E2E=1 and real DATADOG_API_KEY/DATADOG_APP_KEY are set in .env. Example: RUN_E2E=1 npm run test:e2e. Use this to verify that document-center production error logs are visible (e.g. logsDocumentCenter.e2e.test.js).

Operational notes

• Logging: Tool calls are logged to stderr as JSON lines (tool, durationMs, slow). Optional: set MCP_SLOW_TOOL_MS (default 2000) to mark slow calls. Some clients also write to mcp_datadog.log (see src/utils/logger.js).

• Rate limiting: The server does not rate limit; high tool usage can hit Datadog API limits.

• Troubleshooting: Tools missing → check MCP config and env vars, restart client. 403/404 → permissions or plan. See Troubleshooting for "no data" cases.

Troubleshooting

Search returns 0 logs but I expect data

If search_logs (or aggregate_logs) returns no results for a service you know has traffic:

1. Same Datadog org – The MCP server uses DATADOG_API_KEY, DATADOG_APP_KEY, and optionally DATADOG_SITE. Ensure these point to the same Datadog org and site where your app (e.g. document-center) sends logs.

2. Compare in Datadog – In Datadog Logs Explorer, run the same filter and time range (e.g. service:document-center env:production status:error, last 7 days). If you see logs there but not via MCP, the env/site or keys are likely different.

3. Exact filter syntax – Confirm the attribute names and values your app sends (e.g. service, env, status). Try without env:production or with env:prod, or search only service:document-center to see if any logs appear.

4. Retention and indexes – Logs must be in an index that your API key can read; check log indexes and retention.

Documentation

• README (this file) – Setup, usage, structure.

• CLAUDE.md – Project conventions and patterns for contributors.

• docs/ – Additional guides (e.g. performance, security) when present.

Contributing

1. Follow existing style (npm run lint, npm run format).

2. Add tests for new behavior.

3. Run npm run validate before committing.

4. Use conventional commits: feat, fix, docs, chore, refactor, test.

Links

• Datadog API

• Model Context Protocol