What is the Model Context Protocol (MCP)?
Skip this and you rebuild the same CRM connector three times, once for each AI client your team adopted.
Model Context Protocol
MCP
Model Context Protocol (MCP) is an open standard for connecting AI applications to external systems such as files, databases and APIs. A program called an MCP server exposes what it offers. Any MCP client can then use it, with no custom integration written for that client.
Anthropic's documentation compares MCP to a USB-C port for AI applications. The comparison holds up. You wire the port once, and whatever gets plugged in later works.
A server can expose three things, and the spec names them. Tools are functions the AI can call. Resources are data it can read. Prompts are reusable templates. Messages travel as JSON-RPC 2.0 over one of two transports. Stdio runs a server as a local process on the same machine. Streamable HTTP runs it remotely, over the network, with normal HTTP authentication.
The integration cost stops multiplying
Before MCP, connecting five systems to three AI clients meant fifteen integrations. Each client had its own way of describing a tool, so the work did not carry over. MCP turns that into five servers and three clients that already speak the standard.
Support is now broad. Claude, ChatGPT, Visual Studio Code and Cursor all act as MCP hosts, so a server you write for one is usable by the others.
- 01Your engineers write one server per system, not one per system per client.
- 02Tool lists are fetched at runtime, so adding a capability does not need a client release.
- 03Auth stays where it belongs: remote servers use bearer tokens, API keys or OAuth.
- HostThe AI app: Claude, an IDE.
- ClientOne per server, holds the link.
- DiscoverAsks what the server supports.
- Listtools/list returns the schemas.
- Calltools/call runs one, result returns.
Nothing here decides whether a call is allowed. That check belongs in your server, and it is the step teams skip.
Related questions
01Is MCP only for Claude?
No. MCP is an open standard, and the documentation lists Claude, ChatGPT, Visual Studio Code, Cursor and others as clients. Anthropic created it and open-sourced it, which is why the association sticks.
02What is the difference between an MCP server and an API?
An MCP server is a wrapper that describes an API in a form models can discover and call. Your REST endpoint stays where it is. The server publishes a name, a description and a schema for each action. A model can then find the right one with no developer wiring it up first.
03Should we build an MCP server or a normal integration?
Build an MCP server when more than one AI client needs the same system, or when you expect that number to grow. For a single product with one model behind it, plain tool calling is less machinery for the same result.
