You have probably seen “MCP” everywhere lately. Every AI tool seems to support it now. But what is it actually?

The problem before MCP

Imagine you have 10 AI tools and 20 data sources. Databases, APIs, file systems, Slack, GitHub.

Each AI tool needed its own custom integration with each data source. That is 200 different integrations to build and maintain.

Developers called this the N×M problem.

What MCP does

Model Context Protocol is a standard. Like USB-C, but for AI.

It defines one way for any AI model to connect to any tool or data source. Build the integration once. Every AI that supports MCP can use it.

N×M becomes N+M.

Anthropic open-sourced it in November 2024. By 2026, every major AI ships MCP support — Claude, ChatGPT, Gemini, Cursor, GitHub Copilot, VS Code. It crossed 97 million monthly SDK downloads and is now under Linux Foundation governance (donated December 2025).

How it works

There are three parts:

MCP Host — the AI app you use. Claude Desktop, Cursor, your IDE.

MCP Client — lives inside the host. Talks to servers on behalf of the AI.

MCP Server — connects to a real tool. Your database, your GitHub, your file system.

When Claude needs to query your database, it does not do it directly. It asks the MCP client. The client asks the right server. The server runs the query and sends back the result.

What MCP servers can do

There are three types of capabilities:

  • Tools — functions the AI can call. query_database, send_email, read_file
  • Resources — data the AI can read. Files, docs, API responses
  • Prompts — pre-built templates for common tasks

Try it in 5 minutes

Add a filesystem MCP server to Claude Desktop. This lets Claude read and write files on your computer.

Open this file:

~/Library/Application\ Support/Claude/claude_desktop_config.json

Add this (replace the path with your actual projects folder):

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/your-username/projects"
      ]
    }
  }
}

You need Node.js installed for npx to work.

Restart Claude Desktop. Now ask Claude to read a file in your projects folder. It works.

Build your own MCP server

The official Python SDK makes it simple:

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("my-server")

@mcp.tool()
def get_weather(city: str) -> str:
    """Get current weather for a city."""
    # your API call here
    return f"Weather in {city}: 22°C, sunny"

if __name__ == "__main__":
    mcp.run()

That is it. Any MCP-compatible AI can now call your get_weather tool.

Why this matters

Before MCP, every AI coding tool had its own plugin system. Cursor plugins. Copilot extensions. Claude integrations. All different.

Now you build one MCP server. It works everywhere.

This is why it reached 97 million monthly SDK downloads. Not because it is new technology — because it solves a real problem in the simplest way possible.

What’s Next?