What is MCP, and what is an MCP hub?
Updated November 4, 2026
MCP (Model Context Protocol) is an open standard that lets an AI assistant use outside tools: search your mail, read a document, create a task. The assistant is the MCP client; each service is reached through an MCP server. An MCP hub is one MCP server that connects many services and accounts, so you add one address to your AI instead of a server per service.
How MCP works
An MCP server describes its tools to the AI client: a name, what each tool does and which inputs it takes. When you ask for something, the AI picks a tool, the client calls the server, and the result comes back into the conversation.
Local servers run on your computer and talk to the client over standard input and output. Remote servers run in the cloud and are reached by URL over HTTP (the Streamable HTTP transport); you sign in to them with OAuth, so the AI never sees your password.
Which AI apps support MCP
Remote MCP servers work in most AI assistants and coding tools today, including:
- Claude (web, desktop and mobile) and Claude Code
- ChatGPT in developer mode
- Cursor, VS Code with GitHub Copilot, Windsurf
- Codex and Gemini CLI
What an MCP hub adds
With a server per service, every one has to be added and signed in to in every AI app, most allow one account, and the tools of all servers fill the AI’s context. A hub sits between the AI and your services:
- one address for every AI app;
- several accounts per service, each with its own name;
- services without their own MCP server, such as any IMAP mailbox or a REST API;
- the same rules and approvals for every service, and one activity log.
How Manyplug does it
Manyplug is a hosted MCP hub. You sign in with Google, connect accounts on the website or by asking the AI, and paste one address into Claude, ChatGPT or Cursor. 38 services are built in, any other remote MCP server or REST API can be added, keys are encrypted per account, and risky actions such as sending mail wait for your OK by default.