SuperAPI AI Directory

datadog

Deployment & operations

View repository

Updated 10-16 06:38

About this skill

Datadog Observability — Gram Project

Gram Services

Always filter by the relevant service(s) when querying Datadog:

Service Description
gram Dashboard frontend (RUM)
gram-dashboard Dashboard backend
gram-server Main backend API server
gram-worker Temporal worker
fly Fly.io — where Gram Functions logs are collected

Available Tools

Use only the following Datadog MCP tools unless the user explicitly asks for others:

Logs

  • mcp__datadog-mcp__search_datadog_logs — Search and filter log events
  • mcp__datadog-mcp__analyze_datadog_logs — Analyze log patterns and aggregate stats

Metrics

  • mcp__datadog-mcp__get_datadog_metric — Get a specific metric's time-series data
  • mcp__datadog-mcp__get_datadog_metric_context — Get context and metadata for a metric
  • mcp__datadog-mcp__search_datadog_metrics — Search available metrics by name

Traces & Spans

  • mcp__datadog-mcp__get_datadog_trace — Get a specific trace by ID
  • mcp__datadog-mcp__search_datadog_spans — Search spans (useful for latency investigation)

Incidents & Monitors

  • mcp__datadog-mcp__search_datadog_incidents — Search active/recent incidents
  • mcp__datadog-mcp__get_datadog_incident — Get details for a specific incident
  • mcp__datadog-mcp__search_datadog_monitors — Find monitors and their current state

RUM & Events

  • mcp__datadog-mcp__search_datadog_rum_events — Search Real User Monitoring events (frontend errors, sessions)
  • mcp__datadog-mcp__search_datadog_events — Search Datadog events stream

Services & Infrastructure

  • mcp__datadog-mcp__search_datadog_services — Discover services in APM
  • mcp__datadog-mcp__search_datadog_service_dependencies — View service dependency map

Guidelines

  • Always scope queries to one or more Gram services using the service filter when available.
  • Start narrow, expand if needed: Query a 15–30 minute window first, then widen.
  • For error investigations: start with search_datadog_logs, filter by status:error, then follow trace IDs with get_datadog_trace.
  • For latency issues: use search_datadog_spans with service:<name> and sort by duration.
  • For frontend issues: prefer search_datadog_rum_events for gram.
  • For incidents: check search_datadog_incidents first before deep-diving into logs.