In one minute
An AI assistant on its own can only work with what you type or paste into the chat. MCP gives it a standard way to reach further: read files on your computer, look up an issue in your tracker, search documentation, or create a record in your CRM.
The work is split into three roles:
- Host: the AI app you use, such as Claude Desktop, ChatGPT, Cursor or VS Code.
- Client: the part of the host that holds one connection to one server. A host with three servers runs three clients.
- Server: a program that offers capabilities to the host. It can run on your own computer or on the internet (see remote vs local MCP servers).
What a server can offer
The protocol defines three things a server can expose:
- Tools: actions the model can call, such as "search issues" or "create a file". See tool call.
- Resources: data the app can read, such as a file's contents or a database schema.
- Prompts: reusable templates for common tasks.
Most servers today are mainly a set of tools. The app asks a server for its list of tools, shows them to the model, and runs the ones the model chooses, usually after asking you to approve.
Why it matters
Before MCP, every AI app needed its own integration for every service. With MCP, a service builds one server and it works in every app that speaks the protocol. As of 19 November 2026 the MCP site lists Claude, ChatGPT, Visual Studio Code and Cursor among the apps that support it.
The current version of the specification is dated 2026-07-28. Messages use JSON-RPC 2.0, and servers talk to apps over one of two transports: standard input and output for servers on your computer, or Streamable HTTP for servers on the internet.
Where to go next
- Add an MCP server to your assistant, starting with Claude Desktop.
- Build your first MCP server in TypeScript or Python.
- Browse MCP servers in the directory.