Explain / MCP · Model Context Protocol

What is an MCP?

A way to use a website or tool from inside your AI — and let the AI see a bit of what you’re doing there. An API, but inside your assistant.

In plain English

MCP, or Model Context Protocol if you want the full name, is a bridge between your AI and the tools you actually use. It lets your assistant connect to a service you already have, your calendar, a writing tool, a database, read what’s there, and do something with it, all from inside the chat. Think of it as handing the AI a set of keys to your filing cabinet instead of photocopying every page and posting it through the letterbox. You still choose which drawers it can open.

The shortest way to say it: an API, but inside your AI.

A worked example

Say you run a freelance writing business, juggling a calendar full of deadlines and a folder of client notes. Right now you probably copy the deadline into ChatGPT by hand, then paste in your notes too. Bit tedious. With your calendar and notes connected over MCP, you open the chat and it can already see what’s due and what the client asked for. You type “draft something for the Johnson piece due Friday” and it pulls the date, checks the brief, and starts writing, no pasting required.

What you actually see is the AI working from your real information instead of guessing, and a draft that already knows the deadline without you saying it twice.

The usual mix-up

People assume MCP means the AI can wander round your accounts doing whatever it likes. It can’t. It only sees and touches what you’ve connected and allowed, and that’s the whole point of setting it up properly.

Where you’ll use it

Also comes up in: build me a website as a designer, sound like me, sound like my brand.

All the words