MCP Endpoint Inspector
Connect to an MCP server, list its tools and schemas, and see what an agent would run into.
About the MCP Endpoint Inspector
MCP Endpoint Inspector connects to a Model Context Protocol server over HTTP and shows what an AI agent would see: the server's name and protocol version, the capabilities it advertises, and every tool with its description and parameters. It performs the same opening exchange a real client does, so if the inspector can read your server, an agent probably can too.
When something is wrong, the report says what. A URL that redirects, one that only answers GET, a server that returns a web page instead of a JSON-RPC reply, and one that needs a login are each named with the likely cause. When the connection works, it checks the tool list for the things that make tools hard for models to use: missing or very long descriptions, duplicate or oddly named tools, input schemas that are not objects, and parameters nobody described. Agents pick tools by reading these descriptions, so they matter more than they look.
The inspector supports the Streamable HTTP transport and does not send credentials, so a protected server is reported as requiring authentication rather than opened. Use your own client for those. To connect your agent to toolhq's own tools, see the MCP page, or point this inspector at https://toolhq.io/api/mcp to see it working.
Reach the developers and engineers who use these tools every day.
Frequently asked questions
Which MCP transports does it support?
Streamable HTTP, where the client POSTs JSON-RPC messages and the server replies with JSON or an event stream. The older HTTP plus SSE transport and local stdio servers are not supported, because they cannot be reached from a web service.
Why does it say the server requires authentication?
The server answered with a 401 or 403. The inspector deliberately sends no credentials, tokens, or cookies, so it will not look inside a protected server. Inspect those from your own MCP client, which can complete the login.
What does the inspector do to my server?
It sends an initialize request, the initialized notification, and read only list requests for tools, resources, and prompts. It never calls a tool. If your server issues a session, the inspector ends it when it finishes.
Why do descriptions matter so much?
A model decides which tool to call by reading its name, description, and parameter descriptions. A tool with no description is rarely chosen, and a vague one is chosen at the wrong time. Clear, specific descriptions are the cheapest way to improve an MCP server.
Can I inspect a server running on localhost?
No. The request comes from our server, which refuses private and local addresses. Expose the server through a public tunnel first, or use a client that runs on your own machine.