DataDog MCP Server

Evidence-driven MCP server that exposes DataDog as tools, resources, and prompts. Uses the official DataDog API and MCP Python SDK.

Requirements

• Python 3.10+
• DataDog API key and Application key (create here)

Quick start

1. Clone or enter this repo, then install: uv sync
2. Copy .env.example to .env and set DD_API_KEY and DD_APP_KEY (create keys at DataDog API keys). Never commit .env.
3. Run the server: uv run python -m datadog_mcp.server (stdio) or use streamable HTTP (see below).

Configuration

Set credentials via environment variables (do not commit real values):

Variable	Required	Description
DD_API_KEY	Yes	DataDog API key
DD_APP_KEY	Yes	DataDog Application key
DD_SITE	No	Site (default: datadoghq.com; use datadoghq.eu, ap1.datadoghq.com, etc.)

Copy .env.example to .env and fill in values locally; never commit .env.

Cursor / MCP client setup

To attach this server as an MCP in Cursor (or another stdio client):

• Command: uv run python -m datadog_mcp.server (run from the datadog-mcp directory).
• Env: Ensure DD_API_KEY and DD_APP_KEY are set in the environment where the client runs (e.g. in Cursor’s MCP settings or your shell).

Example MCP config snippet (path and env depend on your setup):

{
  "mcpServers": {
    "datadog": {
      "command": "uv",
      "args": ["run", "python", "-m", "datadog_mcp.server"],
      "cwd": "/path/to/datadog-mcp",
      "env": { "DD_API_KEY": "<your-api-key>", "DD_APP_KEY": "<your-app-key>" }
    }
  }
}

Running the server

Default: stdio (for Cursor, CLI clients):

cd datadog-mcp
uv run python -m datadog_mcp.server
# or with explicit stdio:
uv run python -m datadog_mcp.server stdio

Streamable HTTP (for MCP Inspector, browser-based clients):

uv run python -c "from datadog_mcp.server import run; run(transport='streamable-http')"

Then connect the MCP Inspector to the shown URL (e.g. http://localhost:8000/mcp).

Tools

Tool	Description	Scope
validate_keys_tool	Validate API/Application keys	—
list_monitors, get_monitor, create_monitor, update_monitor, delete_monitor, mute_monitor, unmute_monitor	Monitors	monitors_read, monitors_write
list_dashboards, get_dashboard, create_dashboard, update_dashboard, delete_dashboard	Dashboards	dashboards_read, dashboards_write
query_metrics	Metrics query	metrics_read
list_log_indexes, query_logs	Logs	logs_read_config, logs_read
list_events	Events	events_read
list_hosts, get_host_tags	Hosts / tags	infrastructure_read, tags_read
list_slos, get_slo	SLOs	slo_read
list_downtimes, get_downtime, create_downtime, update_downtime, cancel_downtime	Downtimes (v1)	monitors_read, monitors_write
list_incidents, get_incident, create_incident, update_incident	Incidents (v2)	incident_read, incident_write
list_apm_services	APM services	apm_read
list_synthetics_tests, get_synthetics_test	Synthetics	synthetics_read
list_notebooks, get_notebook	Notebooks	notebooks_read
get_usage_summary	Usage metering	usage_read

Resources (URI)

• datadog://validate — API key validation status
• datadog://monitors — List of monitors
• datadog://monitors/{id} — Single monitor by ID
• datadog://dashboards — List of dashboards
• datadog://dashboards/{id} — Single dashboard by ID
• datadog://downtimes — List of downtimes
• datadog://downtimes/{id} — Single downtime by ID
• datadog://slos — List of SLOs
• datadog://slos/{id} — Single SLO by ID
• datadog://incidents — List of incidents
• datadog://incidents/{id} — Single incident by ID

Prompts

• prompt_summarize_monitor_state(monitor_id) — Ask the LLM to summarize a monitor’s state
• prompt_draft_incident_status(incident_id) — Ask the LLM to draft a status message from an incident
• prompt_summarize_slo(slo_id) — Ask the LLM to summarize SLO state and burn rate
• prompt_dashboard_insights(dashboard_id) — Ask the LLM to summarize key widgets and suggest focus areas

Full endpoint → tool mapping and scopes: docs/api_mapping.md. For detailed tool parameters, scopes, and examples: docs/tool_reference.md.

Development

uv sync
uv run pytest

DataDog scopes and rate limits

• Scopes: See Authorization Scopes. Each tool documents the scope it needs; grant the minimum required.
• Rate limits: See Rate Limits. On 429 the server retries with backoff (see docs/tool_reference.md); if still rate-limited, it returns a sanitized error.

Secrets

Do not log, echo, or commit DD_API_KEY or DD_APP_KEY. Use environment variables or a secure secret manager only.