MCP Inspector

MCP Inspector is the client you use while building or debugging an MCP server. Point it at a server and it lists the tools, resources and prompts the running server exposes; you can call each one by hand and read the raw response. Use it to check tool schemas and responses before you add a server to Claude Code or Cursor.

Setup

npx @modelcontextprotocol/inspector opens the web UI on port 6274; --cli and --tui start the command-line and terminal clients. It requires Node 22.19 or newer. Name a server ad hoc (mcp-inspector node build/index.js for stdio, --server-url https://api.example.com/mcp --transport http for remote) or keep a list in ~/.mcp-inspector/mcp.json. The web backend binds to 127.0.0.1 and requires a random per-launch token (MCP_INSPECTOR_API_TOKEN sets it, DANGEROUSLY_OMIT_AUTH disables it).

What it exposes

• A web UI for browsing and calling tools, resources and prompts, OAuth-protected servers included
• CLI methods such as tools/list, tools/call --tool-name x --tool-arg key=value and resources/list
• --format json output for CI, and --strict checks for tool-schema portability
• stdio, SSE and Streamable HTTP transports

Limits

• Without an OS keychain (headless Linux, SSH sessions, containers), OAuth client secrets and stdio env values go to ~/.mcp-inspector/secrets.json in plain text by default.
• v2 changed the CLI flags; v1 only gets security fixes.

Useful for anyone writing an MCP server, and for checking what a third-party server offers before installing it.