The MCP inspector reads a Model Context Protocol config or a tools/list payload and shows servers, commands, environment keys, and each tool's input schema. It runs in your browser.
Cursor, Claude Desktop, and other agents store MCP servers in mcp.json under mcpServers. Each entry is either a local stdio process (command plus args) or a remote URL. A tools/list response is the other half: the tools that server exposes, with JSON Schema for the arguments.
Why use it
See every server, its transport, and the exact command line without opening the JSON by hand.
Check which environment variables a server expects. Values stay hidden until you choose Show env.
Read a tool's required fields before you write a prompt or a client call.
Nothing is uploaded. History and share only keep the text you pasted, so do not share a file that contains live tokens.
How to use
Paste mcp.json or a tools list. The Config and Tools list chips load samples.
For a config, read each server card: stdio command, HTTP URL, and env keys.
For tools, pick a name chip and read the property list plus the raw input schema.
Use Share or History when you want to come back to the same payload.
Config vs tools/list
A config answers "what process will start?" A tools list answers "what can the model call?" They are different documents. If a server fails to connect, check the command and args here, then ask the client for tools/list and paste that to confirm the schema the agent will send.