Datadog MCP Server

The Datadog MCP that answers "why is this happening?" — not just "what's the value?"

Aggregation tools that fold 5–7 sequential API calls into one structured response. Full SLO CRUD. Fleet automation. The widest Datadog API coverage in any MCP — 163 tools built on the @us-all MCP standard.

What it does that others don't

• Aggregation tools — analyze-monitor-state and slo-compliance-snapshot collapse 5–7 sequential API calls into one structured response with a caveats array for partial failures. No other Datadog MCP ships this pattern.
• Full SLO CRUD — create, update, delete SLOs (and their corrections). The official Bits AI MCP and community alternatives are read-only on SLOs.
• Fleet Automation — 15 tools across deployments and schedules. Only this server.
• Status Pages — 21 tools for full status-page lifecycle (components, degradations, maintenances). Only this server.
• Token-efficient by design — extractFields projection, DD_TOOLS/DD_DISABLE 16-category toggles, and a search-tools meta-tool keep LLM context low across 163 tools.
• Apps SDK card — slo-compliance-snapshot renders as a visual card on ChatGPT clients via _meta["openai/outputTemplate"]. Claude clients receive the same JSON content (non-breaking).
• stdio + Streamable HTTP — defaults to stdio (Claude Desktop / Code). Set MCP_TRANSPORT=http for ChatGPT Apps SDK or remote clients (Bearer auth via MCP_HTTP_TOKEN).

Try this — 5 prompts

Connect the server to Claude Desktop or Claude Code, then paste any of these:

1. SLO health — "List my SLOs and their error budget remaining this month. Group by status: compliant, at-risk, breached."
2. Incident triage — "There's an active incident on checkout-service. Pull the linked monitors, the recent error spikes from APM, and which deployments touched the service in the last 24h."
3. Monitor noise audit — "Find monitors that alerted more than 10 times in the last 7 days but had MTTR under 5 minutes — these are probably flapping."
4. RUM error spike — "RUM error rate jumped on the checkout funnel between 14:00 and 14:30 today. Show me the top error groups, affected sessions, and the user actions before the errors."
5. Fleet rollout — "Schedule the datadog-agent 7.55.0 rollout to the staging cluster, weekends only, starting next Saturday."

When to use this vs Datadog's official MCP

Datadog's official MCP (Bits AI MCP, GA 2026-03-09) is complementary, not a replacement:

Use the official Bits AI MCP for fast managed onboarding and SSO. Use this when you need full API coverage, SLO/fleet/status-page write parity, or self-hosting (internal sites, isolated networks, dev/CI sandboxes).

Install

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "datadog": {
      "command": "npx",
      "args": ["-y", "@us-all/datadog-mcp"],
      "env": {
        "DD_API_KEY": "<your-api-key>",
        "DD_APP_KEY": "<your-app-key>",
        "DD_SITE": "datadoghq.com"
      }
    }
  }
}

Claude Code

claude mcp add datadog -s user \
  -e DD_API_KEY=<your-api-key> -e DD_APP_KEY=<your-app-key> -e DD_SITE=datadoghq.com \
  -- npx -y @us-all/datadog-mcp

Docker

docker run -e DD_API_KEY=... -e DD_APP_KEY=... -e DD_SITE=datadoghq.com \
  ghcr.io/us-all/datadog-mcp-server:latest

Build from source

git clone https://github.com/us-all/datadog-mcp-server.git
cd datadog-mcp-server && pnpm install && pnpm build
node dist/index.js

Configuration

Categories (16): metrics, monitors, dashboards, logs, apm, rum, incidents, security, synthetics, ci, infra, fleet, status-pages, oncall, teams, account.

When MCP_TRANSPORT=http: POST /mcp (Bearer-auth JSON-RPC) + GET /health (public liveness).

Sites:

Token efficiency

Naive setup loads ~25K tokens of tool schema before any conversation. Three knobs mitigate:

1. Category toggles — DD_TOOLS=metrics,monitors,logs,apm (biggest win).
2. extractFields response projection — get-dashboard { dashboardId: "abc", extractFields: "id,title,widgets.*.definition.type" }.
3. search-tools meta-tool — always enabled; lets the LLM discover tools at runtime instead of preloading all schemas.

Read-only mode

By default, all writes are blocked to prevent accidental mutations by AI agents. The following require DD_ALLOW_WRITE=true:

create-monitor, update-monitor, delete-monitor, mute-monitor, create-dashboard, update-dashboard, delete-dashboard, send-logs, post-event, trigger-synthetics, create-synthetics-test, update-synthetics-test, delete-synthetics-test, create-downtime, cancel-downtime, create-case, update-case-status, send-dora-deployment, send-dora-incident, create-slo, update-slo, delete-slo, plus all fleet/status-page/security writes.

MCP Prompts (4)

Workflow templates the model can invoke directly:

• triage-incident — given an incident ID, walks linked monitors, recent error spikes, and recent deploys.
• audit-monitor-noise — flag flapping monitors via alert frequency × MTTR.
• analyze-rum-error-spike — diff RUM error rates across two windows, attribute to top error groups.
• investigate-slow-trace — given a slow trace ID, traverse the span tree and surface bottleneck spans.

MCP Resources

Read-only entities by URI: dd://monitor/{id}, dd://dashboard/{id}, dd://slo/{id}, dd://incident/{id}, dd://service/{serviceName}, dd://team/{teamId} (team + members), dd://synthetics/{testId}, dd://host/{name}.

Tool reference

163 tools across 16 categories. Use the search-tools meta-tool to discover at runtime; the full list is collapsed below.

Metrics (5)

query-metrics, get-metrics, get-metric-metadata, list-active-metrics, list-metric-tags

Monitors (7)

get-monitors, get-monitor, create-monitor, update-monitor, delete-monitor, mute-monitor, validate-monitor, analyze-monitor-state (aggregation)

Dashboards (5)

get-dashboards, get-dashboard, create-dashboard, update-dashboard, delete-dashboard

Logs (3)

search-logs, aggregate-logs, send-logs

Events (2)

get-events, post-event

Incidents (6)

get-incidents, get-incident, search-incidents, create-incident, update-incident, delete-incident

APM (1)

search-spans

RUM (17)

search-rum-events, aggregate-rum, list-rum-applications, get-rum-application, create-rum-application, update-rum-application, delete-rum-application, list-rum-metrics, get-rum-metric, create-rum-metric, update-rum-metric, delete-rum-metric, list-rum-retention-filters, get-rum-retention-filter, create-rum-retention-filter, update-rum-retention-filter, delete-rum-retention-filter

SLOs (6)

list-slos, get-slo, get-slo-history, create-slo, update-slo, delete-slo, slo-compliance-snapshot (aggregation), plus 5 SLO-correction tools

Synthetics (6)

list-synthetics, get-synthetics-result, trigger-synthetics, create-synthetics-test, update-synthetics-test, delete-synthetics-test

Hosts / Containers / Processes (4)

list-hosts, get-host-totals, list-containers, list-processes

Downtimes (3)

list-downtimes, create-downtime, cancel-downtime

Security (9)

search-security-signals, get-security-signal, list-security-rules, get-security-rule, delete-security-rule, list-security-suppressions, get-security-suppression, create-security-suppression, delete-security-suppression

CI Visibility (4)

search-ci-pipelines, aggregate-ci-pipelines, search-ci-tests, aggregate-ci-tests

Cases (4)

list-cases, get-case, create-case, update-case-status

Error Tracking (2)

list-error-tracking-issues, get-error-tracking-issue

DORA (2)

send-dora-deployment, send-dora-incident

Network Devices (2)

list-network-devices, get-network-device

Notebooks (2)

list-notebooks, get-notebook

OnCall (2)

get-team-oncall, get-oncall-schedule

Services & Software Catalog (2)

list-services, get-service-definition

Teams (6)

list-teams, get-team, create-team, update-team, delete-team, get-team-members

Account & Users (2)

get-usage-summary, list-users

Logs/Spans/APM Retention metrics (15)

5 each for logs-metrics, spans-metrics, apm-retention-filters (list/get/create/update/delete)

Status Pages (21)

Full lifecycle: pages, components, degradations, maintenances. See src/tools/status-pages.ts.

Fleet Automation (15)

Agents, deployments, schedules. See src/tools/fleet.ts.

Audit (1)

search-audit-logs

Meta (1)

search-tools — query other tools by keyword; always enabled regardless of DD_TOOLS.

Architecture

Claude → MCP stdio → index.ts → tools/*.ts → @datadog/datadog-api-client → Datadog API

Built on @us-all/mcp-toolkit:

• extractFields — token-efficient response projections
• aggregate(fetchers, caveats) — fan-out helper for aggregation tools
• createWrapToolHandler — domain-specific redaction (DD_API_KEY/DD_APP_KEY) + Datadog ApiException error extraction
• search-tools meta-tool

Tech stack

Node.js 22+ • TypeScript strict ESM • pnpm • @modelcontextprotocol/sdk • @datadog/datadog-api-client (official) • zod • dotenv • vitest + dd-trace.

Contributing

See CONTRIBUTING.md. New shared patterns belong in @us-all/mcp-toolkit — single source of truth for the 7-server suite.

License

MIT