Datadog MCP Server
Officialby Datadog
Ask Claude Code 'why did latency spike on the checkout service at 3am?' and get an answer pulled from your live Datadog logs, traces, and metrics -- without opening a browser, without switching to the Datadog dashboard, without writing a single query. Datadog MCP Server went GA on March 9, 2026, making Datadog the first major observability platform to ship a native Model Context Protocol integration. It's a remote MCP server -- no local installation required -- that connects your AI coding agents directly to Datadog's unified observability data. Your agent gets live access to logs, metrics, traces, dashboards, monitors, incidents, hosts, services, events, and notebooks through 16+ core tools. The design philosophy is modular: rather than dumping 50 tools into every agent context and burning tokens, you opt into specific toolsets based on your stack. The core toolset gives you the essentials. Need deeper analysis? Enable the APM toolset for trace analysis, span search, Watchdog insights, and performance investigation. Running feature flags? The feature-flags toolset lets your agent create, list, and update flags across environments. The LLM observability toolset is purpose-built for teams monitoring their own AI systems -- recursive, but practical. Authentication uses your existing Datadog API key and App key. No new credentials, no OAuth dance, no separate auth service. If you have Datadog access, you have MCP access. Four real-world use cases are already emerging from early adopters: onboarding new engineers (agent identifies relevant monitors and dashboards), service decommissioning (agent detects unused services by filtering synthetic traffic from logs), incident correlation (agent cross-references alerts with feature flag changes), and cost anomaly detection (agent monitors cloud cost metrics and flags spikes). The limitation is straightforward: this is a read-heavy server. Most operations are queries and analysis. Write capabilities exist for some toolsets (feature flags, cases), but you're not going to deploy infrastructure or modify monitors through MCP. And you need to be a Datadog customer -- there's no free tier for the MCP Server alone.
Installation
Key Features
- ✓16+ core tools covering logs, metrics, traces, dashboards, monitors, incidents, hosts, services, events, and notebooks
- ✓Modular toolsets: APM (trace analysis, Watchdog insights), Error Tracking, Feature Flags, DBM, Security, LLM Observability
- ✓Remote server -- no local installation, connects via Datadog API key and App key
- ✓Compatible with Claude Code, Cursor, GitHub Copilot, VS Code, Codex, Cognition, and Kiro
- ✓Feature flag management: create, list, and update flags across environments from your IDE
- ✓Case management with Jira linking for incident workflows
- ✓Alerting toolset for monitor validation, group search, and template management
- ✓Database Monitoring toolset for query performance analysis and anomaly detection
Use Cases
- →Debug production incidents without leaving your IDE: 'show me error logs from the payments service in the last hour' returns live Datadog data
- →Onboard new engineers faster: agent identifies relevant monitors, dashboards, and service maps for their team's infrastructure
- →Correlate incidents with deployments: 'did any feature flags change before the latency spike on checkout-service?' cross-references signals automatically
- →Detect cost anomalies: agent monitors cloud spend metrics and alerts when daily costs exceed baseline by more than 20%
- →Service decommissioning: agent analyzes log traffic patterns to identify truly unused services vs. those with only synthetic health-check traffic