generated: '2026-09-05'
method: searched
source: >-
  https://docs.datadoghq.com/bits_ai/mcp_server/ (fetched as .md),
  https://docs.datadoghq.com/mcp_server/tools.md, https://docs.datadoghq.com/mcp_server/setup.md,
  https://github.com/datadog-labs/opencode-plugin (first-party endpoint literal), and live probes of
  mcp.datadoghq.com
provider: Datadog APM
providerId: datadog-apm
name: Datadog MCP Server
status: published
description: >-
  Datadog operates a first-party remote MCP server that bridges AI agents to Datadog observability
  data. It is not APM-specific — it exposes most Datadog products — but it ships a dedicated `apm`
  toolset for span search, trace queries, Watchdog stories, latency-bottleneck analysis and APM
  recommendations, which is the agent-facing counterpart of the REST surface captured in openapi/.
deployment:
  mode: remote
  endpoint: https://mcp.datadoghq.com/api/unstable/mcp-server/mcp
  auth: oauth
  verified: probed
  note: >-
    Endpoint taken verbatim from Datadog's own opencode-plugin README, then probed: it answers
    HTTP 401 {"errors":["Unauthorized"]} to an anonymous tools/list, and mcp.datadoghq.com publishes
    RFC 8414 + RFC 9728 metadata for it. The docs pages themselves render the endpoint from a
    client-side site selector, so the literal is sourced from the first-party repo rather than the
    rendered page. No stdio package is published for this server; the vendor plugins (Claude Code,
    Copilot, OpenCode, Cursor) are wrappers that point a client at this same remote URL.
transport: streamable-http
regional_endpoints:
  note: >-
    Datadog is a multi-site product and the docs present a per-site endpoint through a JavaScript
    site selector. Only the US1 endpoint above was observed directly; the other site endpoints were
    not fetched and are deliberately not listed rather than pattern-guessed.
authorization:
  model: oauth2
  spec_version: '2.1'
  issuer: https://mcp.datadoghq.com/v1/mcp
  authorization_endpoint: https://app.datadoghq.com/oauth2/v1/authorize
  token_endpoint: https://app.datadoghq.com/api/v2/oauth2/token
  registration_endpoint: https://app.datadoghq.com/api/v2/oauth2/register
  pkce_required: true
  code_challenge_methods:
  - S256
  scopes:
  - mcp_all
  bearer_methods:
  - header
  source: https://mcp.datadoghq.com/.well-known/oauth-authorization-server
  evidence: well-known/datadog-apm-oauth-authorization-server.json
probe:
  url: https://mcp.datadoghq.com/api/unstable/mcp-server/mcp
  method: POST
  body: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
  http_status: 401
  response: '{"errors":["Unauthorized"]}'
  fetched: '2026-09-05'
  conclusion: >-
    The live tool schemas are auth-gated. The tool list below is therefore transcribed from Datadog's
    published tools reference (names, toolset, required permission, description) — inputSchema for
    each tool requires authenticated introspection and is deliberately absent rather than invented.
authorization_model_note: >-
  The MCP Server forwards the authenticated user's own credentials to Datadog APIs, so existing RBAC,
  Data Access Control and log restriction queries apply unchanged; write tools check the matching
  permission (for example monitors_write) on each call.
rate_limits:
  burst: 50 requests / 10 seconds
  monthly: 100000 tool calls
  source: https://docs.datadoghq.com/bits_ai/mcp_server/
  note: Datadog labels these fair-use limits and says they are subject to change.
toolsets:
  selector: '?toolsets=<name>[,<name>] query parameter (or the X-Datadog-MCP-Toolsets header for Codex)'
  all: toolsets=all
  omit: '?omit_tools=<name>'
  apm_toolset_status: Preview — access requires signing up at https://www.datadoghq.com/product-preview/apm-mcp-toolset/
tools:
- name: apm_search_spans
  toolset: apm
  permission: APM Read
  description: Searches for spans using APM query syntax, with support for pagination and tag filtering.
- name: apm_query_trace
  toolset: apm
  permission: APM Read
  description: >-
    Queries a trace's span data to filter, aggregate, or rank spans, such as finding the highest
    self-time spans or tracing an error to its origin service.
- name: apm_discover_span_tags
  toolset: apm
  permission: APM Read
  description: Discovers available tag keys on spans within a time range.
- name: apm_get_primary_tag_keys
  toolset: apm
  permission: APM Read
  description: Retrieves the primary tag keys configured for the organization.
- name: apm_search_watchdog_stories
  toolset: apm
  permission: APM Read
  description: >-
    Searches for Watchdog anomaly detection stories for a service within a time range, providing
    AI-powered insights into latency, error rate, and traffic anomalies.
- name: apm_get_watchdog_story
  toolset: apm
  permission: APM Read
  description: Retrieves detailed information about a specific Watchdog story by its ID.
- name: apm_latency_bottleneck_summary
  toolset: apm
  permission: APM Read
  description: >-
    Analyzes latency bottlenecks across traces in an anomaly period using self-time calculations,
    detects cascading call patterns, and surfaces root causes of latency spikes.
- name: get_change_stories
  toolset: apm
  description: >-
    Retrieves change stories from the Change Tracking API for APM services — deployments, feature
    flags, configuration updates and infrastructure events in a time range.
- name: semantic_search_change_stories
  toolset: apm
  description: Searches change stories using natural language and AI-powered semantic search.
- name: apm_search_recommendations
  toolset: apm
  permission: APM Read
  description: Searches for APM recommendations from Datadog.
- name: apm_get_recommendation
  toolset: apm
  permission: APM Read
  description: Retrieves full details of a specific APM recommendation by ID.
- name: get_datadog_trace
  toolset: core
  description: Core-toolset trace retrieval, available without the apm preview toolset.
- name: search_datadog_spans
  toolset: core
  description: Core-toolset span search.
- name: search_datadog_services
  toolset: core
  description: Core-toolset service search.
- name: search_datadog_service_dependencies
  toolset: core
  description: Core-toolset service dependency search.
- name: search_datadog_slos
  toolset: alerting
  permission: SLOs Read
  description: Searches Datadog SLOs by name, tags, or type, with query syntax for service/team filtering.
tools_note: >-
  Only the APM-relevant tools are transcribed here. Datadog's published reference lists well over a
  hundred tools across ~20 toolsets (logs, metrics, dashboards, cases, DDSQL, database monitoring,
  error tracking, security, synthetics and more); the full list is at
  https://docs.datadoghq.com/mcp_server/tools.md.
observability:
  metrics:
  - datadog.mcp.session.starts
  - datadog.mcp.tool.usage
  audit_trail: All tool calls are recorded in Datadog Audit Trail under the event name "MCP Server".
compliance:
  hipaa_eligible: true
  govcloud_supported: false
docs:
  overview: https://docs.datadoghq.com/bits_ai/mcp_server/
  setup: https://docs.datadoghq.com/mcp_server/setup/
  tools: https://docs.datadoghq.com/mcp_server/tools/
related_servers:
- name: Datadog Code Security MCP Server
  mode: local-stdio
  scope: >-
    A separate, narrower local-only server for SAST, SCA, IaC, secrets and SBOM scanning against
    source code. Not the APM surface; recorded so the two are not conflated.
  docs: https://docs.datadoghq.com/security/code_security/dev_tool_int/mcp_server/
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com