What is MCP? The standard for connecting AI to your tools
What the Model Context Protocol is, how it works under the hood and how to connect your first MCP server to Claude Code in a few minutes.
On this page
Before you start: for the hands-on example you need Claude Code installed. If you don't have it yet, follow Install Claude Code step by step.
If you use an AI assistant to write code, you've probably hit this wall: the model is very capable, but it can't see your tools. It doesn't know what's in your database, your task tracker or your team's internal docs. MCP exists to solve exactly that.
MCP in one sentence
MCP (Model Context Protocol) is an open standard that lets AI applications connect to external tools and data sources in a common way.
Before MCP, every integration was custom work: one connector for each pair of assistant and tool. With MCP, a tool exposes an MCP server once, and any compatible application can use it. It's the same idea as a USB port: one connector, many devices.
The three pieces
MCP splits the connection into three roles:
- Host: the AI application you use, such as Claude Code, Claude Desktop or a code editor.
- Client: the component inside the host that keeps the connection to each server.
- Server: a small program that exposes a tool's capabilities, such as your database, your repository or your ticketing system.
What a server can offer
An MCP server exposes three kinds of capabilities:
- Tools: actions the model can run, such as
create_taskorfind_customer. - Resources: data the model can read, such as files, records or documentation.
- Prompts: reusable instruction templates the server makes available to you.
The model decides when to use each tool, but you approve sensitive actions before they run.
How they talk
The client and the server exchange JSON-RPC messages over one of two common transports:
- stdio: the server runs as a local process and talks over standard input and output. This is the most common setup for tools on your machine.
- HTTP: the server runs remotely and connects over the network. This is the usual setup for cloud services.
Your first MCP server in Claude Code
Imagine your team uses a fictional internal notes app called NotesLab, and someone has already published its MCP server. Connecting it to Claude Code takes a single command:
claude mcp add noteslab -- npx -y @noteslab/mcp-server
To check which servers you have connected:
claude mcp list
From there you can ask Claude things like "search NotesLab for the architecture decisions on project Atlas", and it will use the server's tools to answer with real data instead of guesses.
NotesLab, Atlas and the
@noteslab/mcp-serverpackage are fictional examples used to illustrate the flow.
Mind the security
An MCP server can read data and take actions on your behalf, so treat it like any other dependency in your project:
- Install servers only from sources you trust.
- Check which permissions it needs and grant the minimum.
- Read carefully the actions the agent asks you to approve.
Wrapping up
MCP turns AI from an isolated assistant into one that works with your real tools. One standard, many servers, and the same way to connect them in any compatible application.
In upcoming videos, Khez and KJ build an MCP server from scratch, step by step.