Lesson 19 of 28 · Making AI useful

What is the Model Context Protocol (MCP)?

MCP stands for Model Context Protocol, an open standard that defines how AI applications connect to outside tools and data, such as files, calendars, databases or business apps. Anthropic introduced it in November 2024. A tool is wrapped once as an MCP server, and any AI app that supports MCP can then use it, much as many devices share one USB-C socket.

The problem MCP solves

An AI model on its own can only work with what is in its prompt. To check your calendar, search your company wiki or update a ticket, the AI app needs a connection to those systems. Before MCP, every app had to build its own connector for every tool, so ten AI apps and ten tools could mean a hundred separate integrations, each built and maintained differently.

A shared protocol, meaning an agreed set of rules for how two programs talk, changes the maths. Each tool is connected once, each app learns the protocol once, and they all work together. It is the same idea that lets any website open in any browser.

How MCP works, in plain terms

MCP has three roles. The host is the AI application you use, such as a chat assistant or coding tool. Inside it, a client handles the connection to each MCP server. The server is a small program that wraps one tool or data source and describes what it offers.

When you ask something, the host shows the model what the connected servers offer; the model asks to use one, and the host carries out the request and passes back the result. A server can offer three kinds of things:

Tools: actions the model can request, such as 'create a calendar event'.
Resources: information it can read, such as a file or a database record.
Prompts: ready-made templates for common tasks.

What MCP is not

MCP is not an AI model and not an agent. It is plumbing: a standard way for an AI app or agent to reach tools. It is not a product you buy either. Although Anthropic introduced it, it is an open standard with a public specification, designed to work with any model or app.

MCP is also not a safety guarantee. It standardises the connection, not what is connected. A server can expose private data or powerful actions, so install servers only from sources you trust, grant the narrowest permissions, and be careful when a connected tool returns content from the open web, which can contain hidden instructions.

Where you meet MCP

In AI assistants and coding tools, MCP usually shows up as 'connectors', 'integrations' or 'MCP servers' in the settings, where you can add access to a service such as a file store, issue tracker or database. Agent frameworks and workflow tools use it to give agents a standard toolbox, and software vendors publish MCP servers so their product can be used from AI apps.

When choosing AI tools, MCP support is a quick signal of how easily a product connects to the rest of your software. For how servers are built, the transports they use and how to judge one, read the detailed guide to MCP servers.

Example

Worked example: one question, one MCP round trip

Worked example

Setup: your AI assistant (the host) is connected to a calendar MCP server. You ask: 'Do I have anything on Friday afternoon?' 1. The host tells the model the calendar server offers a tool: list_events(date). 2. The model asks to call list_events(date: Friday). 3. The host sends the request to the calendar server, which checks your calendar. 4. The server returns: 2 pm Design review; 4 pm Call with supplier. 5. The model answers: 'Yes, a design review at 2 pm and a supplier call at 4 pm.'

Illustrative flow with a made-up tool name. Many hosts ask for your approval before a tool runs, especially for actions that change data.

Key points

MCP stands for Model Context Protocol, an open standard introduced in November 2024.
It lets AI apps connect to tools and data through one shared method.
A tool is wrapped once as an MCP server and works with any MCP-compatible app.
MCP is plumbing, not a model or an agent.
It standardises connections, not safety: install only trusted servers.

Try it on Mindber

Go deeper: what is an MCP server?

Glossary: Model Context Protocol

Search the directory for MCP tools

AI agents topic hub

AI agents

Frequently asked questions

Is MCP only for Claude?

No. Anthropic introduced MCP and uses it in its own products, but it is an open standard with a public specification, and any AI application or model provider can implement it. Developers build MCP servers for all kinds of services, and a server written once can work with any host that supports the protocol, whichever model sits behind it.

What is the difference between MCP and an API?

An API is how one program exposes its functions to others, and every API has its own design. MCP is a standard layer on top: an MCP server often calls a service's existing API, but presents it to AI apps in one common format, with descriptions the model can understand. That way, an AI app does not need custom code for each API.

Do I need to be a developer to use MCP?

Not to use it. In many AI assistants you add an MCP server from a settings screen or a list of connectors, then simply ask questions that need it. Building your own MCP server is a developer task, helped by the official software development kits. Either way, review what access a server requests before you connect it.