Claude, ChatGPT, Claude Code and Cursor then reach it through the same MCPifex gateway as every other server, and you choose the tools each key may call.

Start free

Why build a custom MCP server?

Because the API you need has no MCP server. Many internal tools and smaller SaaS products have a good REST API and no MCP server, so Claude and ChatGPT cannot use them. Writing one yourself means coding a tool for each endpoint, hosting it, keeping the API key safe and updating it when the API changes. Connect any API does that work from the docs you already have.

How does Connect any API work?

  1. Name the service or paste its docs URL. Type a name such as Resend, or paste the link to the API docs.
  2. MCPifex reads the API. The scan looks for an OpenAPI, Swagger or Postman spec first and falls back to the HTML docs. It finds the base URL, where the key goes (header, Bearer, query or Basic), extra settings such as a region or account id, and every endpoint. Each endpoint becomes a tool marked read or write.
  3. Add your API key and run Test connection. Test connection makes one safe read call with your key. You cannot save until it passes.
  4. Pick the tools for your first key. Read tools start on and write tools start off. Change them, then create the key.
  5. Paste one URL into your AI app. Add https://mcp.mcpifex.com/mcp and your mcpx_ key to Claude, ChatGPT, Claude Code or Cursor.

The URL is the same for every server you connect:

https://mcp.mcpifex.com/mcp

The new key only sees the tools you picked. You can add more keys later, each with its own tools, and revoke one without touching the others. API keys and authentication covers both ways to send the key.

What do I get?

  • A private server. The server is yours or your team's. Nobody else sees it in the catalog.
  • Tools set per key. Give your assistant a read-only key and your developer a key with writes, on the same API.
  • Writes start off. Endpoints that create, change or delete start switched off for every key.
  • Rescans keep your choices. When the API changes, rescan it. Your tool choices stay, and new tools arrive off.
  • One instance. A connected API counts as one instance on your plan. Free covers 3.
  • Same endpoint as every server. It sits next to PostgreSQL, Search Console and the other hosted servers behind one gateway URL.

The tool rules work the same way as on the gateway. Read choose which tools each key can use for why it is set per key. If you are weighing an API call against an MCP tool, see MCP vs API and what MCP tools are.

How is this different from OpenAPI-to-MCP generators?

Generators turn an OpenAPI spec into MCP server code, and you still run that code. You host it, store the API key, update it when the spec changes and decide who may call which tool. Connect any API reads the spec or the docs, hosts the server, keeps the key in your MCPifex instance and adds per-key tool control. If you want to own the code, a generator is the better fit. If you want the API in Claude today, connect it here.

What are the limits?

  • API-key authentication only. APIs that need OAuth cannot be connected as custom servers.
  • The scan is only as good as the docs. Test connection must pass before you save, and you can rescan later.
  • Your API key is not sent to the scanner. It is used for Test connection and for your AI's calls.
  • Each connected API counts as one instance. Free covers 3 instances, Pro 10 and Team 30. Compare them on pricing.

Prefer a ready-made server? The marketplace has hosted servers for PostgreSQL, Google Search Console, Bing Webmaster Tools and more.

Connect any API FAQ

  • A custom MCP server exposes your own API, or a service that has no official MCP server, as tools an AI app can call. With Connect any API, MCPifex builds and hosts that server for you from the API's spec or docs.

Ready to try it?

Run hosted MCP servers behind one endpoint and choose exactly which tools your AI can use.