English
Workflow collection · 2026-10-18

Observability & Monitoring MCP Servers and Skills

Move from a symptom to supporting evidence: inspect service metrics, narrow the time window, examine logs and traces, and carry the findings into an incident or evaluation workflow. These guides cover both connected telemetry tools and instructions for investigating them.

15 guides · 7 MCP / 8 Skills

Guides are grouped by the evidence they help you inspect. Existing Grafana, Sentry, Dynatrace, Datadog, Instana, and VictoriaMetrics guides sit alongside the new SigNoz, Axiom, Honeycomb, Rootly, MLflow, and W&B entries. Skills and MCP servers remain separate guide types.

Choose your starting point

Your taskStart hereHow to choose
Check service healthMetrics, logs, and service healthChoose the telemetry system already receiving your service’s data and establish a time window.
Investigate an application errorErrors and investigation guidanceUse issue or trace context to form a hypothesis, then test it against a bounded telemetry query.
Prepare an incident handoffRootly MCPCarry incident context and unresolved follow-ups into a handoff; avoid inferring causes that telemetry has not established.
Evaluate an AI applicationAI traces and evaluationUse model and agent trace context together with labeled evaluation data rather than relying only on infrastructure health.

Metrics, logs, and service health

Start with connected operational telemetry and narrow the affected service and time window.

Errors and investigation guidance

Connect application issues with investigation methods and operational context.

AI traces and evaluation

Inspect agent traces, model experiments, application behavior, and evaluation evidence.

SkillBatch 2026-10-16

LangSmith Skills

Inspect LLM application traces, build evaluation datasets, create evaluators, and work with LangSmith custom apps using publisher-authored skills.

langchain-ai/langsmith-skillsDescription source checked 2026-10-15
Read directory guide →Author source ↗

Questions about this collection

Will installing a skill give it access to my telemetry?

Instructions alone do not grant account access. Some agent plugins also configure an MCP connection, while script helpers can require separate credentials. Follow the individual guide’s setup notes.

How should I move between these guides during an investigation?

Keep the same environment, service, and time window. Start with service-level evidence, then inspect representative errors or traces. Carry links and unresolved questions into your incident handoff.

Are infrastructure monitoring and AI evaluation interchangeable?

They answer different questions. Metrics and logs help explain service behavior; agent traces and evaluations help examine model and workflow quality. Use the guide that matches the evidence you need.

Keep exploring

Explore the fourth batch: 29 official cloud, payment, security, and monitoring additions →

See all 25 additions in the third batch → · Full directory →