What is MCP Inspector?
MCP Inspector is the official interactive developer tool for testing and debugging MCP servers, run via npx @modelcontextprotocol/inspector. It connects to a server and provides a UI for browsing and exercising its tools, resources, and prompts.
WHY IT MATTERS
The Inspector runs without installation: npx @modelcontextprotocol/inspector <command> wraps any server launch command — an npm package via npx, a Python package via uvx, or a local build via node path/to/index.js. A connection pane lets you choose the transport and, for local servers, customise arguments and environment variables.
Once connected, it surfaces everything the server declares:
- Tools tab — lists available tools with their schemas and descriptions, and lets you invoke them with custom inputs and inspect the results.
- Resources tab — lists resources with metadata and content, including subscription testing.
- Prompts tab — shows prompt templates and previews generated messages with custom arguments.
- Notifications pane — streams logs and notifications received from the server.
It is the standard tool in two situations. During server development, the loop is: change code, rebuild, reconnect the Inspector, re-test affected features — including edge cases like invalid inputs and error responses. Before adopting a third-party server, it is the quickest way to see exactly what tools/list returns and how the server behaves, rather than relying on its README. It inspects one server interactively at a time, so it complements rather than replaces automated security scanning across a fleet.
HOW POLICYLAYER USES THIS
PolicyLayer does at fleet scale what the Inspector does interactively for one server: its crawler discovers servers, extracts their tool definitions, and publishes per-tool risk classifications in the catalogue at policylayer.com/tools. Once a server is vetted and registered, the gateway enforces policy on every call to it — inspection tells you what a server exposes; enforcement decides what actually runs.