Free tool

Claude Desktop MCP Config Validator

Paste your config and see what Claude Desktop will actually do with it. Most broken MCP configs are valid JSON — which is exactly why nothing tells you they are wrong.

Paste the JSON config. Remove any token first if you would rather — the checks do not need it.
Runs entirely in your browser. Nothing you paste is uploaded, stored or logged.

Connected at last. Now show someone what it does.

Getting the config right is the boring half. MCP Showcase takes the same server and produces a live playground your customers can try, with documentation for every tool.

Start for Free Explore the MCP server directory

How it works

link
Paste your config

Nothing is uploaded. The check runs entirely in your browser.

play_circle
We read it as Claude Desktop would

Root key, endpoint key, transport type and the client-specific fields that Claude Desktop treats differently from everyone else.

checklist
Fix what it finds

Each problem comes with what Claude Desktop expects instead, so it is a change you can make rather than a symptom to search for.

Why a valid config still does nothing

Almost every broken MCP config is valid JSON. Clients ignore keys they do not recognise, so a file with the wrong root key or the wrong endpoint field parses cleanly, Claude Desktop starts normally, and no servers appear. There is no error to search for. That silence is why these mistakes survive so long.

What this checks for Claude Desktop

  • Root key — Claude Desktop reads mcpServers. Anything else is parsed and ignored.
  • Endpoint field — Claude Desktop cannot take a hosted URL at all; a remote server has to be bridged through a local command.
  • Entry shape — every server needs either a command to launch or an endpoint to connect to.
  • Plain HTTP — flagged, because any bearer token sent to that server travels unencrypted.
  • Claude Desktop-specific fields — see below.

Remote servers need a local bridge

Claude Desktop's config launches local processes; it does not take a hosted MCP URL directly. A remote server is configured as a stdio command that runs the mcp-remote proxy and is given the URL as an argument. Pasting the URL into a "url" field is the single most common mistake here, and it fails silently -- the server simply never appears.

Your config is not uploaded

The check runs entirely in your browser and makes no network request. That is deliberate rather than incidental: MCP configs routinely contain API tokens, and "paste your config into this website" is a reasonable thing to refuse. You can confirm it in your browser's network tab.

If the config is right but the server is empty

Then the problem has moved out of the file. A client that shows the server but no tools is connecting and failing at the handshake or the transport, not misreading your config. The MCP Inspector shows which transport the server answers on and the raw request log, and the MCP Server Tester reduces the same thing to a pass/fail checklist.

Related tools

Once it connects, the hard part starts

A working config is table stakes. What it does not do is help anyone understand what your server is for — and if you publish an MCP server, that is the part that decides whether anyone configures it at all. MCP Showcase turns the same endpoint into a live playground with per-tool documentation, so evaluating it takes a click instead of a config file and a restart.

Frequently asked questions

Because clients ignore what they do not recognise. A config with the wrong root key, or a remote endpoint under the wrong field name, parses perfectly and results in a client that sees no servers and reports nothing. Silence is the normal failure mode here.

That the root key is the one Claude Desktop reads, that remote servers use the field name Claude Desktop expects, that each entry has either a command or an endpoint, that nothing is served over plain HTTP, and the client-specific quirks that apply to Claude Desktop in particular.

No. It is parsed in your browser and never leaves your machine — which matters, because MCP configs frequently contain API tokens. No network request is made at all.

It depends entirely on the client, which is why this catches so many people. Claude, Cursor, Windsurf and Cline use mcpServers; VS Code uses servers; Zed uses context_servers; Codex CLI uses mcp_servers in TOML. The result below names the one Claude Desktop needs.

Then the problem is at the server rather than in the file. Run the endpoint through the MCP Inspector to confirm it completes a handshake and to see what it advertises.

More free MCP tools

MCP Inspector Online

Paste an MCP server URL and see every tool, resource and prompt it exposes, with full schemas and the raw request log. Nothing to install.
MCP Server Tester

Check that your MCP server actually works. Paste the URL and get a pass/fail report on the connection, the handshake, the transport and the tools it advertises.
MCP Security Scanner

Before you point an agent at an MCP server, see what it is asking for. Graded A–F on transport, authentication, tool-poisoning markers and dangerous parameters.
MCP Token Calculator

Tool definitions are sent on every request, not once. Paste your MCP server URL to see how many tokens that is, which tools are the heaviest, and what it costs.
See all free tools