What is MCP?

Open the settings in many chatbots and you'll find a list of connectors: your calendar, your files, a notes app, a work tool, each with a Connect button. Press one, approve it, and the assistant can see or do things in that app. Behind a lot of those buttons sits one shared standard. It's called the Model Context Protocol (MCP). You don't need to know how it works to press those buttons, but knowing a little helps you choose them well.

Before it, each connection was built by hand. If an artificial intelligence (AI) app wanted to read your calendar, someone wrote custom code for that calendar. Another app wanted the same calendar? Someone wrote it again. Ten apps and ten tools could mean a hundred separate links, each one built and kept working on its own. Anthropic, the company that makes Claude, named the problem when it introduced the standard: every new data source needed its own custom work, which made connected systems hard to scale.¹

MCP replaces those one-off links with one shared way of connecting. Anthropic released it as an open standard in November 2024, so anyone can build with it. Make one MCP connection for a tool, and any app that speaks MCP can use it. It doesn't matter which company made the app or the tool. That's the whole idea: a common plug, so an assistant can reach many tools the same way, and a tool only has to be wired up once.¹²⁴

There are two sides. On one side is the app you talk to, such as a chatbot or a coding assistant. On the other is a small program called a server, one for each tool, that knows how to talk to it: a calendar server, a files server, a database server. The server says what its tool offers, and the app asks for what it needs. Neither side has to know how the other was built. They only have to speak the same protocol, which here just means an agreed set of rules for the conversation.¹⁴

A plug only helps if people trust it to stay open. In December 2025, Anthropic handed MCP to a new group, the Agentic AI Foundation, run under the Linux Foundation, a nonprofit that looks after many shared software projects. Anthropic co-founded it with Block and OpenAI, with support from Google, Microsoft, and others. So the standard is now looked after jointly, and no one company owns the plug.³

The standard decides how things connect. You decide what gets connected. Some connections can only read, like looking at your calendar. Others can act: send a message, change a file, make a booking. Security guidance for these systems says the same thing in technical terms: give each connection only the access its job needs. When you press Connect, read what it asks for. If you stop using one, disconnect it. And before the next Connect button, ask yourself two plain things: what would this let it see, and what would it let it do?⁵