MCP Inspector

MCP Inspector is the debugging tool you'll open dozens of times a day while building MCP servers. It's a web UI that lets you manually invoke tools, browse resources, and test prompts without touching a full client like Claude Desktop. I keep it running in a browser tab whenever I'm iterating on server code. The workflow is dead simple: run your server through Inspector's npx wrapper, and it spins up a local web interface. You get a clean three-panel layout showing available tools, resources, and prompts. Click a tool, fill in the arguments through a generated form, hit invoke, see the response. No config files, no API keys, no fuss. It's faster than restarting Claude Desktop and navigating through chat to test a single function. The resource browser is particularly useful. I can see exactly what my server exposes, click through to fetch content, and verify URIs match what I expect. The prompt testing panel works the same way, though I find myself using it less since most servers focus on tools and resources. MCP support is complete for the features it claims: tools, resources, and prompts all work. Sampling is marked false, which makes sense since Inspector is for manual testing, not autonomous agent behaviour. The UI shows request/response JSON, so you can debug serialisation issues or unexpected parameter shapes. Rough edges: the interface is functional but basic. No request history, no saved test cases, no diff view between responses. If your tool returns large JSON blobs, you're scrolling through raw output. The npx wrapper occasionally trips over complex server startup commands, though you can work around it by pointing Inspector at a running server instead. I wouldn't use Inspector as my primary MCP client for actual work. It's a developer tool, not a chat interface. But for the 'does this tool actually work' question, it's faster and more transparent than any alternative. Every MCP server author will end up here eventually.