Datadog Mcp

Datadog MCP Server

[Quality gate] [CI/Release] [npm] [License] [Coverage]

DISCLAIMER: This is a community-maintained project and is not officially affiliated with, endorsed by, or supported by Datadog, Inc. This MCP server utilizes the Datadog API but is developed independently.

MCP server providing AI assistants with full Datadog observability access. Features grep-like log search, APM trace filtering with duration/status/error queries, smart sampling modes for token efficiency, and cross-correlation between logs, traces, and metrics. Supports both stdio (local) and http (remote/Kubernetes) transports.

Quick Start

Minimal Claude Desktop / VS Code / Cursor config — just the two required keys:

{  "mcpServers": {    "datadog": {      "command": "npx",      "args": ["-y", "datadog-mcp"],      "env": {        "DD_API_KEY": "your-api-key",        "DD_APP_KEY": "your-app-key"      }    }  }}

With optional tuning (EU site, custom default limits, longer log windows):

{  "mcpServers": {    "datadog": {      "command": "npx",      "args": ["-y", "datadog-mcp"],      "env": {        "DD_API_KEY": "your-api-key",        "DD_APP_KEY": "your-app-key",        "DD_SITE": "datadoghq.eu",        "MCP_DEFAULT_LIMIT": "50",        "MCP_DEFAULT_LOG_LINES": "200",        "MCP_DEFAULT_METRIC_POINTS": "1000",        "MCP_DEFAULT_TIME_RANGE": "24"      }    }  }}

To run as an HTTP server (e.g. inside a container or Kubernetes pod), add transport variables to the same env block:

"env": {  "DD_API_KEY": "your-api-key",  "DD_APP_KEY": "your-app-key",  "MCP_TRANSPORT": "http",  "MCP_PORT": "3000",  "MCP_HOST": "0.0.0.0"}

Configuration

Required environment variables

DD_API_KEY=your-api-keyDD_APP_KEY=your-app-key

Optional environment variables

DD_SITE=datadoghq.com  # Default. Use datadoghq.eu for EU, etc.
# Limit defaults (fallbacks when the AI doesn't specify)MCP_DEFAULT_LIMIT=50              # General tools default limitMCP_DEFAULT_LOG_LINES=200         # Logs tool default limitMCP_DEFAULT_METRIC_POINTS=1000    # Metrics timeseries data pointsMCP_DEFAULT_TIME_RANGE=24         # Default time range in hours
# Transport (alternative to CLI flags — useful in Kubernetes)MCP_TRANSPORT=stdio               # stdio | httpMCP_PORT=3000                     # HTTP portMCP_HOST=0.0.0.0                  # HTTP host

Optional flags

--site=datadoghq.com     # Datadog site (overrides DD_SITE)--transport=stdio|http   # Transport mode (default: stdio)--port=3000              # HTTP port when using http transport--host=0.0.0.0           # HTTP host when using http transport--read-only              # Block all write operations--disable-tools=synthetics,rum,security    # Comma-separated list of tools to disable

Transports

Select with --transport=http or MCP_TRANSPORT=http.

Deployment

Docker

{  "mcpServers": {    "datadog": {      "command": "docker",      "args": [        "run", "-i", "--rm",        "-e", "DD_API_KEY",        "-e", "DD_APP_KEY",        "-e", "DD_SITE",        "ghcr.io/tantiope/datadog-mcp"      ],      "env": {        "DD_API_KEY": "your-api-key",        "DD_APP_KEY": "your-app-key",        "DD_SITE": "datadoghq.com"      }    }  }}

Kubernetes

Use environment variables — not container args — for transport configuration:

env:  - name: DD_API_KEY    value: "your-api-key"  - name: DD_APP_KEY    value: "your-app-key"  - name: MCP_TRANSPORT    value: "http"  - name: MCP_PORT    value: "3000"  - name: MCP_HOST    value: "0.0.0.0"

Note: Kubernetes args: replaces the entire Dockerfile CMD, causing Node.js to receive the flags instead of your application. Environment variables avoid this issue.

Tools

Limit Control

AI assistants have full control over query limits. The MCP_DEFAULT_* environment variables only set the fallback used when the AI doesn't specify a limit — they do NOT cap what the AI can request.

Tool-level token reduction features (compact: true on logs, sample: "diverse" | "spread" | "first", field projections, diagnostics) are surfaced in each tool's MCP description and chosen by the AI at call time.

Notable behaviors

A handful of patterns worth knowing about — the AI can discover the rest from tool descriptions.

• Renotifies vs real fires. monitors top and events search with source:alert count every renotify Datadog emits (one every renotify_interval while a monitor is Alert). To get actual state transitions, use monitors history (defaults to transitionType: ["alert","alert recovery"]) or pass transitionType to events search.
• DST-safe time buckets. events histogram buckets by hour_of_day / day_of_week / day_of_month in any IANA timezone via Intl.DateTimeFormat. Cursor-paginates the underlying search; cap controlled by MCP_MAX_EVENTS_HISTOGRAM (default 5000) with bucketCountIncomplete + nextCursor on overflow.
• Validate before create. monitors create with dry_run: true calls /api/v1/monitor/validate instead of persisting. Allowed in --read-only mode.
• Monitor template preview. monitors preview renders a notification against a context payload — variable substitution + Datadog's six documented conditionals (is_alert, is_warning, is_no_data, is_recovery, is_alert_to_warning, is_warning_to_alert) + the tag conditionals {{#is_match "tag" "val"}} (substring) and {{#is_exact_match "tag" "val"}} (exact), with ^ negations and OR'd multiple comparison values (resolved against context.variables; case-sensitive). {{#each}} and partials throw EUNSUPPORTED_TEMPLATE_SYNTAX.
• SLO round-trip. slos get projects query, monitorIds, monitorTags, groups, and a UI url so you can edit and feed back into slos update without dropping definition fields.
• Cross-correlation. logs(sample:"diverse") → pull dd.trace_id → traces(query:"trace_id:<id>") → metrics(query:"p95:trace.express.request{service:...}") (root metric without .duration for percentiles).

Deep links

Every query response includes a datadog_url field built for your configured DD_SITE — datadoghq.com (default), .eu, us3 / us5 / ap1.datadoghq.com, or ddog-gov.com. Supported on logs, metrics, traces, events, monitors, rum, slos.

Contributing

Contributions are welcome! Feel free to open an issue or a pull request if you have any suggestions, bug reports, or improvements to propose.

License

This project is licensed under the Apache License, Version 2.0.