we_are_coded.by CODE · The world, decoded
БГ
Concept

MCP (the common plug)

The BasicsUpdated on 16 August 2026we are coded

The common plug through which the AI assistant connects to your mail, your calendar, your files, instead of every connection being wired by hand.

Checked on16 August 2026
In short: MCP (Model Context Protocol) is an open protocol introduced by Anthropic in November 2024 that describes how an AI model connects to outside programs and data in one standard way. Put simply: a common plug through which the AI assistant connects to the programs you work with - mail, calendar, files. Before it, every such connection was made separately and by hand, like a cable custom-built for just one socket. With MCP the connection is the same, no matter what you're actually plugging in. That's why you run into it everywhere lately - it's what decides which AI can touch what at all.

Say you want your AI assistant to check your mail and then write something into your calendar on its own. It sounds simple, but behind that simplicity sits a choice: which code talks to the mail, which talks to the calendar, and whether the two even understand each other. Before MCP, every such pair of connections was written separately, from scratch, for every combination.

MCP is an agreement reached to get out of that mess. Instead of every connection being different, they all go through the same socket. Your mail tells the AI, through MCP, what it can do. Your calendar tells it, through MCP, what it can do. The AI assistant plugs into the same socket both times, without knowing the details of each program in advance.

In practice this means that when you ask an assistant to organize your week, it doesn't write random code on the fly to reach your calendar. It plugs into a socket the calendar has already provided. Swap your calendar for another one tomorrow, and the assistant can still plug in, as long as the new one uses the same plug.

And here's the other side of it, the one rarely talked about. The plug decides not just what the AI assistant CAN do, but also what it CANNOT do. If your mail hasn't granted access to deleting messages through that plug, the assistant physically can't do it, even if you ask. The socket limits exactly as much as it connects.

MCP doesn't make the AI-driven assistant smarter, it just decides what it can reach at all.

Makes you look again

Every day I connect different programs to each other, and before MCP that work was tedious and manual - every new connection meant new special-case code I had to keep from breaking. Now I put the same socket everywhere and save hours that used to go into patching things up.

That's why I see MCP not as a technical detail but as stage direction. It decides what's on stage and what stays behind the curtain - which AI sees what of your life. The more people run through the same socket, the more it matters that the socket itself is built carefully, because everything else passes through it.

The visual is generated code art. No third-party images.
Official primary sources
→Anthropic: Introducing the Model Context Protocol (25 November 2024)→Model Context Protocol: the specification (official site)