What is MCP, the Model Context Protocol?

An open standard for connecting AI applications to tools and data. A system offered once as an MCP server works in any application that speaks MCP.

beginner · Also called MCP

Without a shared standard, every AI application needs its own connector for every system it talks to. Link 5 assistants to 5 systems and someone builds 25 connectors. The Model Context Protocol, MCP for short, is an open-source standard that replaces them with one shape: a system offers its tools once, as an MCP server, and any application that speaks MCP can use them. Its own documentation compares it to a USB-C port for AI applications.

The 3 parts

Host. The AI application, such as Claude Code, Claude Desktop or Visual Studio Code.
Client. A connection the host opens for each server it talks to.
Server. A program that offers data and actions. It can run on your own machine, talking through standard input and output, or remotely, over Streamable HTTP.

The messages between them use JSON-RPC 2.0, a plain format for requests and replies.

What a server offers

Tools: actions the model can ask for, such as searching an issue tracker or creating a calendar event. To the model they are tools like any other, each a name, a description and the inputs it takes.
Resources: data the application can read, such as files or database records.
Prompts: ready-made templates for common jobs.

Claude, ChatGPT, Visual Studio Code and Cursor are among the many applications that can connect to MCP servers.

Where it fits in an agent

An MCP server does not make a model smarter. It gives the agent more tool cards. The model still writes a request for a tool, and the host passes it to the server and brings the result back. On the Claude API, the MCP connector lets a request name remote MCP servers directly, with no client code of your own. It handles tool calls only and is in beta.

What to watch

Every tool a server offers is a card that can be sent with every request. Connect a few servers with many tools each and their cards can fill much of the context before the agent does any work. Tool search, which loads a tool's card only when it is needed, is one answer.

A server is also code that acts with your access. Connect only servers you trust, give each the narrowest access it needs, and ask a person before tools that send, pay or delete. What a server returns is outside content, and it can carry a prompt injection like any web page.

Sources

Where each fact comes from

Tool Use
Giving a model a list of things it can ask to have done, such as search the web or read a file. The model asks; the program around it does the work and hands back the result.
Agent
A model that uses tools in a loop, deciding each next step from what the last one returned, until the task is done.
Function Calling
The developer's name for tool use. The model replies with the name of a function and its arguments in a fixed format, and your code runs it.
Prompt Injection
An attack that hides instructions in text a model reads, such as a web page, an email or a file, so the model follows the attacker instead of you.
Context Engineering
Choosing what goes into a model's context at each step of a task, what to fetch only when needed, and what to clear, note down or summarise as the context grows.