This server automatically adds OpenTelemetry instrumentation to your Python AI agents without manual code changes. It exposes two tools: instrument_agent reads your agent file, applies the ioa-observe-sdk tracing, and writes it back with a backup, while check_instrumentation audits what's missing. It works with LangGraph, LlamaIndex, CrewAI, and raw OpenAI SDK code. Under the hood it uses an LLM (Claude by default, but supports OpenAI, Gemini, Groq, or local Ollama) to intelligently insert the instrumentation calls. After running it, you spin up the included OTel Collector and ClickHouse stack to query spans and trace your agent's execution. Reach for this when you need observability on existing agents but don't want to manually wire up tracing calls.
claude mcp add --transport stdio alanzha2-observe-instrument-mcp uvx observe-instrument-mcp