How it works
Paste a public MCP server URL and click Connect. The server-side proxy validates the address, tries Streamable HTTP and then SSE, and returns the server's tools, resources, and prompts. Calls are live, so review a tool's schema before running it.
Who it's for
This is for developers testing remote MCP servers before wiring them into Claude, ChatGPT, Cursor, or another host. It supports optional bearer authentication, but not interactive OAuth, stdio, localhost, or private-network endpoints.
Public HTTP or HTTPS endpoints only. Tool calls are live and are not sandboxed.
WHAT IS A MCP INSPECTOR?
An MCP inspector is a debugging client that discovers and exercises the tools, resources, and prompts exposed by a Model Context Protocol server. Paste a public endpoint above and this inspector connects through a server-side proxy, shows the server's capabilities, and lets you send real requests without installing Node, npx, or Docker.
New to the protocol? Start with what an MCP server is, then browse the Model Context Protocol hub for guides, projects, and tools.
HOW TO TEST AN MCP SERVER ONLINE
To test a remote MCP server, enter its public HTTP or HTTPS endpoint, add a bearer token only if required, and select Connect. The inspector tries the current Streamable HTTP transport first and falls back to SSE, then lists up to 200 tools, resources, templates, and prompts reported by the server.
- Confirm the server name, version, and advertised capabilities.
- Open a tool and review its input schema before entering arguments.
- Call the tool, read a resource, or fetch a prompt and inspect the raw result.
- Repeat with error cases before adding the server to an AI client.
MCP INSPECTOR PRIVACY AND SECURITY
This inspector does not persist server URLs, bearer tokens, tool arguments, or results in application storage, but it is not a purely client-side tool. Each action passes through this site's server-side proxy and then reaches the MCP server you entered. Analytics record the inspector action and timing, not the token or request payload.
Treat every tool call as a real remote action, not a sandbox. Use a test server or read-only, least-privilege token, inspect the schema before calling a tool, and rotate any credential you suspect was exposed. Private, loopback, and link-local addresses are blocked, so use the official local inspector for stdio or localhost servers.
BUILD AND DEBUG MCP SERVERS
A useful inspector only proves the server behavior you actually exercise. Test success responses, invalid arguments, authentication failures, large outputs, and operations that can change data before you connect the server to a production AI host.
I built this tool while debugging my image generation MCP server and YouTube analysis MCP server. For the protocol changes that make remote inspection easier, read my breakdown of the 2026 stateless MCP specification.
