How to Connect Claude Code to Datadog with MCP

Connecting Claude Code to Datadog lets Claude search logs, metrics, monitors, incidents and traces while it debugs your code. Datadog's MCP Server is remote and uses your own Datadog permissions. Datadog documents a Claude Code plugin and a manual command.

Checked against: Datadog's MCP Server setup page and Claude Code's MCP documentation on 28 October 2026. eSEOspace did not run these steps; menus can change, so follow the vendor's page if it differs.

Before you start

You need Claude Code installed and a Datadog account. Datadog lists two kinds of access:

• User role permissions. mcp_read for tools that read data and mcp_write for tools that create or change resources, plus the normal permission for the resource. Datadog says the Standard Role has both MCP permissions by default.
• Organization settings. An admin must enable MCP Access. MCP Write Access controls whether write tools appear at all.

Find your Datadog site. Datadog's endpoints by site:

Datadog says the MCP Server isn't supported on its government sites.

Steps

Option A: Datadog plugin (recommended)

Datadog says the plugin "packages the Datadog MCP Server with bundled skills and auto-updates." If you added the server by hand before, remove it first to avoid conflicts.

1. In Claude Code, install the plugin:

/plugin install datadog@claude-plugins-official

1. Run /ddsetup, or just ask a Datadog question. Select your Datadog site and complete the OAuth login.
2. Run /ddtoolsets to turn groups of product-specific tools on or off.
3. After any configuration change, run /reload-plugins and sign in again by opening /plugin and selecting the Datadog plugin.
4. Check you have the permissions for the resources you want.

The plugin repository lists more commands, such as /ddconfig to change your site or organization. See Claude Code plugins and marketplaces.

Option B: Manual setup

If the plugin isn't available to you, Datadog gives this command. Replace the placeholder with your site's endpoint from the table:

claude mcp add --transport http datadog-mcp <YOUR_MCP_SERVER_ENDPOINT>

For US1, that becomes claude mcp add --transport http datadog-mcp https://mcp.datadoghq.com/v1/mcp. Then run /mcp in Claude Code to complete OAuth, as Claude Code's docs describe for remote servers.

To add toolsets, append them to the URL. Datadog's example loads only APM and LLM Observability tools:

<YOUR_MCP_SERVER_ENDPOINT>?toolsets=apm,llmobs

If your organization signs in through a custom subdomain, add ?subdomain=<SUBDOMAIN>. Combine parameters with &.

Header-based sign-in

Datadog says OAuth is recommended, and headers are for when you can't complete OAuth, such as CI. It prefers a Personal Access Token or Service Access Token sent as Authorization: Bearer <YOUR_ACCESS_TOKEN>. An API key and application key sent as DD_API_KEY and DD_APPLICATION_KEY headers also work.

Check that it works

Run claude mcp list and look for ✔ Connected, or open /mcp in a session.

Example prompts:

• "Search Datadog logs for errors from the checkout service in the last hour." (uses search_datadog_logs)
• "List Datadog monitors that are currently alerting." (uses search_datadog_monitors)
• "Show recent Datadog incidents and summarize the open ones." (uses search_datadog_incidents)

Check the tools list for which toolset each tool belongs to.

Permissions and safety

• Your permissions apply. Datadog says the server "cannot grant a user access beyond what that user already has." Write tools check permissions such as monitors_write on each call.
• Org-wide off switch. Turning off MCP Write Access hides write tools for everyone.
• Narrow the tool list. Load only the toolsets you need, and drop single tools with omit_tools.
• Audit Trail. Datadog records every tool call in Audit Trail under the event name MCP Server.
• Data handling. Datadog says the server doesn't send your data to a third-party AI provider; your AI client decides what goes to its model. Use RBAC, Data Access Control and log restriction queries to limit what the server returns.
• Approvals. In Claude Code, keep write tools such as create_datadog_monitor on ask with permission rules.

Logs often hold customer data and secrets. Read data exfiltration through MCP and the MCP Security Center.

If it doesn't work

Datadog's troubleshooting covers missing write tools:

• Organization write access is off. An admin enables MCP Write Access in Organization Settings.
• Role permissions. Your role needs mcp_write plus the resource permission, such as Monitors Write.
• Toolset not enabled. Without toolsets, only core tools load. create_datadog_monitor is in the alerting toolset. Preview toolsets aren't included in toolsets=all.
• Omitted tools. Check the tool isn't listed in omit_tools.

Refresh the tool list after any change; reconnecting from /mcp is a typical way. Other cases:

• Plugin and manual server conflict. Remove the manual server before using the plugin.
• Signed in on the wrong domain. Add the subdomain parameter if your org uses one.
• Rate limited. Datadog lists 50 requests per 10 seconds and 100,000 tool calls per month, "subject to change."

Next step

Read the Datadog MCP server profile for data handling and compliance notes. More setups are in the connection guides hub. For error tracking alongside metrics, see Claude Code with Sentry.

Frequently asked questions