Datadog MCP Server for Claude

Content

Datadog MCP Server is Datadog's official MCP interface for connecting AI agents to Datadog observability and operations context. It lets MCP clients such as Claude, Claude Code, Codex, Cursor, Gemini CLI, Devin, Goose, OpenCode, VS Code, and custom agents query and work with Datadog data through documented MCP tools.

Use it when the assistant needs current Datadog telemetry during debugging, incident response, service review, alert triage, or production-readiness work. Keep the scope tight: start with read-only investigation, enable only the needed toolsets, and require human review before any assistant creates, modifies, publishes, or escalates operational state.

Features

• Official Datadog MCP Server documented under Datadog's MCP Server docs.
• Remote MCP setup for clients that support HTTP transport and OAuth.
• Recommended Datadog Connector path for Claude.
• Recommended Datadog plugin path for Claude Code, with bundled skills and Datadog slash commands.
• Codex setup through ~/.codex/config.toml plus codex mcp login datadog.
• Toolset selection through the toolsets query parameter.
• Tool exclusion through the omit_tools query parameter.
• Core tools for logs, metrics, traces, dashboards, monitors, incidents, hosts, services, events, notebooks, RUM events, and service dependencies.
• Product-specific toolsets for areas such as APM, LLM Observability, alerting, RUM, dashboards, incidents, notebooks, cloud cost, and other Datadog product surfaces as documented by Datadog.
• Context controls such as response truncation and max_tokens parameters for many tools.
• Datadog Audit Trail events and MCP usage metrics for tracking tool calls.

Use Cases

• Ask Claude to summarize active incidents, related monitors, recent deploy events, service dependencies, and likely next read-only checks.
• Investigate production errors by searching Datadog logs, traces, spans, services, dashboards, and relevant metrics from one MCP client conversation.
• Review an alert by pulling monitor context, metric history, service ownership, and linked notebooks before proposing a change.
• Use the APM and LLM Observability toolsets for focused service or model workflow debugging without exposing every Datadog tool.
• Generate a draft monitor or notebook section, then review it in Datadog before publishing or using it in an incident.
• Audit MCP usage through Datadog Audit Trail and MCP usage metrics after a debugging session.

Installation

Claude

Datadog recommends installing the Datadog Connector from the Claude Connectors Directory when it is available. Complete the OAuth flow and verify Datadog permissions for the resources the assistant should access.

Claude Code

Datadog recommends the official Claude Code plugin when available:

/plugin install datadog@claude-plugins-official

For first-time setup, run /ddsetup or enter a Datadog-related prompt, then select the Datadog site, complete OAuth, and use /ddtoolsets to enable or disable product-specific tool groups.

If the plugin is not available, add the remote MCP endpoint directly:

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

Codex

Add the Datadog endpoint for the account's regional Datadog site to Codex:

[mcp_servers.datadog]
url = "<YOUR_MCP_SERVER_ENDPOINT>"

Then authenticate:

codex mcp login datadog

Toolsets

Limit tool access at connection time. For example, this enables only APM and LLM Observability tools:

<YOUR_MCP_SERVER_ENDPOINT>?toolsets=apm,llmobs

Use toolsets=all only after reviewing permissions, client behavior, and context impact. Use omit_tools when a narrower allowlist is not practical.

Configuration

{
  "mcpServers": {
    "datadog": {
      "type": "http",
      "url": "<YOUR_MCP_SERVER_ENDPOINT>?toolsets=apm,llmobs"
    }
  }
}

Examples

Investigate an incident

Use Datadog MCP to list active incidents, related alerting monitors, recent deployment events, and affected services. Do not make changes.

Scope toolsets before debugging

Show which Datadog MCP tools are enabled in this client and what each can read or change before calling any production-impacting tool.

Review observability coverage

Use Datadog MCP to review monitor and SLO coverage for service:checkout. Summarize gaps and propose draft monitor changes without publishing them.

Audit tool usage

Find Datadog Audit Trail entries for MCP Server tool calls from the last debugging session and group them by user, client, and tool name.

Source Notes

• Datadog's MCP Server overview describes the server as a bridge between Datadog observability data and AI agents that support MCP.
• Datadog's AI-agent note says this is the main Datadog MCP Server for most setup questions and distinguishes it from the local Code Security MCP Server.
• Datadog setup docs describe Claude Connector, Claude Code plugin, direct Claude Code HTTP setup, Codex config, OAuth, API/application key fallback, role permissions, toolsets, and omit_tools.
• Datadog tools docs list core tools for events, incidents, metrics, monitors, traces, dashboards, notebooks, hosts, services, logs, RUM events, and related product-specific toolsets.
• Datadog documents MCP Audit Trail records, MCP usage metrics, fair-use rate limits, HIPAA eligibility, no GovCloud compatibility, and 120-day retention for certain remote MCP usage data.

Duplicate Check

Checked current upstream/main, open PR titles, open PR changed files, source URLs, and content files for Datadog MCP, datadog-mcp-server, docs.datadoghq.com/mcp_server, mcp.datadoghq, datadog.mcp, and Datadog Code Security MCP overlap. Existing content mentions Datadog only as an example integration in broad observability or security entries; there is no dedicated Datadog MCP Server entry, source URL duplicate, or open content PR for this server.

Editorial Disclosure

Datadog is a commercial observability and security platform, but this listing is not sponsored, paid, affiliate-backed, or submitted by Datadog. Use Datadog's current documentation, account permissions, regional site selector, privacy terms, and compliance requirements as the source of truth before connecting production data to any AI client.