Datadog MCP Server: Tools, Setup and Security

The Datadog MCP Server lets an AI agent search logs, metrics, traces, hosts and incidents in Datadog, and, with the right permissions, create or change monitors, notebooks and other objects. Datadog publishes and hosts it, with a separate endpoint for each Datadog site.

Checked against: Datadog MCP Server documentation on 28 October 2026. eSEOspace did not install or test this server.

At a glance

What it can do

Datadog's tools reference lists each tool with its toolset and required permission. Tools in the core toolset include:

• search_datadog_logs for log search.
• search_datadog_metrics to list metrics.
• search_datadog_hosts to find monitored hosts.
• search_datadog_incidents to list incidents by state and severity.
• search_datadog_spans and aggregate_spans for APM traces.
• search_datadog_entities to search the service catalog and dependencies.

Product-specific toolsets add more, for example notebooks, security, database monitoring, error tracking and Agent Observability. Write examples in the reference include create_datadog_monitor, create_datadog_notebook, edit_datadog_notebook and delete_datadog_notebook. Datadog notes notebook deletes are soft deletes you can undo within 30 days with recover_datadog_notebook.

You choose toolsets with a toolsets query parameter on the endpoint, such as ?toolsets=apm,llmobs. Datadog says toolsets=all enables "all generally available toolsets." Some toolsets are not generally available. As of October 2026, the reference marks the apm toolset and some cases tools as Preview, and says the investigator toolset, which triggers Bits Investigation runs, is in Preview and needs Datadog support to enable it.

How to connect it

Endpoints by site, from Datadog's setup page:

Claude Code plugin. Datadog recommends its plugin from the official Anthropic marketplace:

/plugin install datadog@claude-plugins-official

Then run /ddsetup, pick your site and complete OAuth. /ddtoolsets turns toolset groups on or off.

Manual setup. Point your client at your site's endpoint, for example:

claude mcp add --transport http datadog-mcp https://mcp.datadoghq.com/v1/mcp

For header-based auth, Datadog prefers a Personal Access Token or Service Access Token sent as a bearer token. API and application keys sent as DD_API_KEY and DD_APPLICATION_KEY headers also work. Setup tabs also cover Cursor, VS Code, Codex, ChatGPT, Claude, Kiro and others. The Datadog ChatGPT app is in Preview for US1 only.

Security notes

• Your permissions apply. Access follows each user's existing Datadog permissions, checked on every tool call.
• Narrow toolsets. Enable only the toolsets you need. You can also drop single tools with omit_tools.
• Audit Trail. Datadog records tool calls in Audit Trail under the event name "MCP Server."
• Data handling. Datadog says the server does not send Datadog data to a third-party AI provider and makes no outbound calls to external domains. Some tools use models hosted by Datadog's AI providers. Generative AI can be disabled per organization through support.
• Retention and limits. Usage data is kept for 120 days. Fair-use limits are 50 requests per 10 seconds and 100,000 tool calls per month, "subject to change."
• Compliance scope. Datadog lists the server as HIPAA-eligible but not GovCloud compatible.

Logs often hold customer data and secrets. Anything the agent reads lands in its context window. Review data exfiltration through MCP and the MCP security hub.

Alternatives

• Grafana MCP server for Grafana, Prometheus and Loki stacks.
• PagerDuty MCP Server for incident response and on-call.
• Cloudflare MCP Server for edge analytics through Cloudflare's GraphQL API.

Next step

Start with the core toolset and a Service Access Token tied to a read-only role. Watch Audit Trail for a week before adding write toolsets. Our MCP readiness audit can help plan that. See more in the server directory.

Frequently asked questions