inspector

Why we included this project

Debugging MCP servers means staring at JSON-RPC traffic, and this tool turns that into a structured workflow. You can open a hosted web app or run it locally to chat with an LLM against your server while watching full traces of tool calls and agent steps, explore the exposed tools, resources, and prompts, and walk through OAuth exchanges with guided checks. It also ships a CLI and an SDK for probing servers, running doctor checks, and asserting on tool and resource shapes from your own tests, so you can push those into CI and run end-to-end tests and spec conformance on every pull request. The evals feature goes beyond debugging: define test cases with expected tool calls, run them across multiple models, and track accuracy over time to catch regressions early. That mix of visibility, programmatic testing, and regression tracking makes it a practical everyday tool for teams building MCP infrastructure.