Connectors Guide

Explainer, written 30 October 2026

What is an MCP connector?

An MCP connector links your AI assistant to another app, so it can read from it and act in it for you: find the note, file the ticket, look up the payment. MCP is the Model Context Protocol, the open standard most assistants now use for this.

How it works

  1. The app runs an MCP server: a service at an address such as https://mcp.example.com/mcp. It offers a short list of tools, each a thing the assistant may do (search notes, create a ticket), with a description the model reads.
  2. You add it to your assistant: in Claude, Customize > Connectors; in ChatGPT, an app in Settings. You sign in to the app, usually with OAuth, so the connector acts as you and sees only what you can see.
  3. The assistant uses the tools when it helps: you ask "what is overdue?", it calls the tool, reads the answer and replies. You can see each call, and write actions can ask for your approval.

Remote and local

A remote connector lives on the internet and works in Claude on the web, desktop and phone, and in ChatGPT. A local MCP server runs on your own computer, and only desktop and developer tools (Claude Desktop, Claude Code, editors) can start it. This directory lists remote ones, because those are what most people can add.

What people connect

Is it safe?

A connector can do what its tools allow, as you. Add connectors from makers you trust, read what a write action will do before you approve it, and remove the ones you no longer use. On Team and Enterprise plans an admin decides which connectors are allowed.

Questions

What is an MCP connector?

An MCP connector links an AI assistant such as Claude or ChatGPT to another app, so the assistant can read from it and act in it for you. It is an MCP server, a small service that speaks the Model Context Protocol, added to your assistant.

Is an MCP connector the same as an MCP server?

The server is the service on the internet; the connector is that server added to your assistant, with your sign-in. People use the words for the same thing.

Do I need to code to use one?

No. In Claude or ChatGPT you pick it from a list, or paste its address, and sign in. Coding is only needed to build one.

Can a connector see everything?

It sees what the account you signed in with can see, and only acts when the assistant uses one of its tools. You can switch it off per chat and remove it at any time.

Next: how to add a connector to Claude, or MCP vs API.