MCP Inspector is a developer tool for inspecting and debugging Model Context Protocol (MCP) servers. One package provides four clients for connecting to a server, listing what it exposes, and calling it: a web app, a scriptable CLI, an interactive terminal UI, and an experimental connection CLI called mcpdo.
The web client is a React single-page app with a Node backend. The CLI targets automation, CI jobs, and fast agent feedback loops: it connects, lists, calls, and asserts, with JSON output and a defined exit-code map, and it keeps OAuth non-interactive. The mcpdo client connects once and then runs many commands against a named connection through an implicit local daemon. All clients share one core library.
The Inspector manages secrets such as OAuth tokens, OAuth client secrets, and stdio environment values. It stores them in the OS keychain by default and falls back to an encrypted or unencrypted file, or memory only, on machines without a keychain. It also runs as a container image and ships end-user agent skills, including one that teaches an agent to drive mcpdo.
Features
- Web UI: a React and Mantine single-page app with a Node backend for interactive server inspection
- CLI: a scriptable client for automation, CI smoke tests, and agent feedback loops, with JSON output and exit codes
- Terminal UI: an interactive TUI built with Ink
- mcpdo (experimental): connects once and runs many commands against a named connection through a local daemon
- Secret storage: keeps OAuth tokens, client secrets, and stdio environment values in the OS keychain, with file or in-memory fallbacks
- MCP App review: a CLI-first, then one-shot web workflow for automated review of MCP App tools
- Test servers: composable MCP test servers and a showcase config that exercises every feature
- Agent skills: end-user skills that teach an agent to drive the mcpdo CLI
- Container image: a Docker image for running the Inspector with durable secrets in a volume
