The State of MCP observability: Observable.tools
summary
Argues that MCP observability is solvable today by marrying MCP with OpenTelemetry/OTLP, demonstrating a concrete method: extract and inject trace context into MCP meta payloads so remote tool spans stitch into the same trace. Shows a working demo sending TypeScript and Python agent traces into W&B Weave and surfacing server-side spans, reports MCP.Run adding OTLP export and observable.tools initiative for vendor-neutral conventions, and recounts OPUS 4 discovering a support bot that autonomously fixed a telemetry mismatch.
AI-written summary · about
chapters (11)
- 0:00Introduction to MCP observability
- 1:07Why MCP observability matters
- 3:55Introducing Observable.tools
- 4:38OpenTelemetry basics
- 6:44Distributed tracing with MCP
- 8:30Merging MCP and OpenTelemetry
- 10:16Demo of MCP traces in Weave
- 11:40Meta MCP server demo
- 13:25MCP.Run’s OpenTelemetry support
- 14:07Call to action and community involvement
- 16:17Closing and resources
related talks
25mLet Me Be Your OpenTelemetry Champ
23mObserving the Observers: Bringing OpenTelemetry to Autonomous AI Agents
26mExploring Observability with MCP Servers
20mObservability and Open Telemetry
55mCloud-Native Observability with OpenTelemetry
37mObservability For You And Me With Opentelemetry
34mLLMs, Meet Your Tools: The Magic of Model Context Protocol
46mThe State of OpenTelemetry