What is MCP? An explanation for people who don't code

You've started seeing three letters in your AI tools. A "connectors" menu in ChatGPT. An MCP setting in Claude. A colleague who says their setup "runs over MCP" with the particular satisfaction of someone who configured a thing.

MCP stands for Model Context Protocol, an open standard published in November 2024 by Anthropic, the company behind Claude. It lets an AI tool connect to outside sources of information and abilities. You set up a connection once, in a settings screen, and every chat can use it. Company Brain holds a team's approved company facts and delivers them to AI tools through MCP.

I run a consultancy alongside building a software product, and I've had to explain MCP across a lot of tables where nobody codes. Here's the version that works.

What MCP is: the socket in the wall

Think about the power socket in your wall. Your lamp doesn't need to know anything about the power plant. The vacuum cleaner doesn't either. There's one standard plug shape, and anything with that plug gets power from anything that provides the socket.

Before the standard existed, every appliance would have needed its own wiring straight to the source. Nobody would buy a lamp.

MCP is that socket, for AI. It's a standard way for an AI tool to connect to outside sources of information and abilities. Anthropic published it, OpenAI adopted it for ChatGPT in March 2025, and Google followed a month later. That last part matters. A plug shape only becomes a standard when everyone uses it.

What MCP means in practice

An AI model on its own knows what it learned in training. General knowledge, frozen at a point in time, with nothing about you in it. Every capability beyond that has to come from a connection.

MCP is how those connections get made. Your calendar becomes something the AI can check. Your project tool becomes something it can read. Your company's approved facts become something it can pull into any conversation, which is the connection we care most about at Company Brain, since it's the one we build. How Company Brain works with Claude shows that setup end to end.

For you, the whole experience is a settings screen. You add a connection once, approve it, and from then on the AI has that source available in every chat. No pasting, no uploading the same file again, no developer in sight.

Why MCP is worth caring about, even a little

For years, the honest answer to "how do I give the AI access to our stuff" was "you build something custom, per tool". A connection to ChatGPT helped nobody using Claude. Whatever you wired up for one tool, you rebuilt for the next.

The standard changes the economics of that. One connection, every compatible tool. When your team switches tools, or uses four at once like most teams quietly do, the connections come along.

So when you see those three letters in a settings menu now, you know what you're looking at. A socket. And the interesting question stops being technical: what should your AI be plugged into?

For most teams I meet, the honest answer starts with their own company's facts. That's the subject of how to stop repeating your company context in every AI chat.