top of page
Search

How to Build Your First MCP Server with Claude Desktop

  • Jun 9
  • 3 min read

MCP (Model Context Protocol) is the new standard that lets Claude directly connect to your local code — giving it the ability to run functions, call APIs, control your computer, and more. In this tutorial, you'll build your first MCP server from scratch using Python and connect it to Claude Desktop.



🎥 Prefer video? Watch the full walkthrough on YouTube → [Your First MCP Server — Codersarts]



What is MCP?

MCP is a protocol that acts as a bridge between Claude Desktop and your local Python code. Once connected, Claude can call functions you define — directly from the chat window — without you having to run anything manually each time.


Think of it like giving Claude a set of custom tools that live on your machine.



What You'll Build

By the end of this tutorial you'll have:

  • A working MCP server with two tools

  • Claude Desktop connected to your local server

  • A clear understanding of how input and output flow between Claude and your code



Prerequisites

  • Python 3.10+

  • Claude Desktop installed (download here)

  • Basic Python knowledge



Step 1 — Install the MCP Package


pip install mcp


Step 2 — Write Your Server


Create a file called server.py:



"""My First MCP Server"""
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Hello World")

@mcp.tool()
def say_hello() -> str:
    """Say hello to the world."""
    return "Hello, World!"

@mcp.tool()
def get_weather(city: str) -> str:
    """Get the weather for a city."""
    return f"Weather in {city}: Sunny, 28°C"

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


What each part does


FastMCP("Hello World") — creates your server. The string is just a label.

@mcp.tool() — registers the function as a tool Claude can call. Without this decorator, Claude has no idea the function exists.


Docstring — Claude reads this to decide when and how to use the tool. Always write a clear one-line description.


mcp.run() — starts the server. The if name == "__main__" guard ensures it only runs when called directly, not when imported.




Step 3 — Configure Claude Desktop

This is where most beginners get confused. You don't run server.py manually — Claude Desktop handles that for you via a config file.



Open Claude Desktop → Settings → Developer → Edit Config

This opens a file called claude_desktop_config.json. Add your server:



{
  "mcpServers": {
    "my-tools": {
      "command": "python",
      "args": ["/absolute/path/to/server.py"]
    }
  }
}



Understanding the config

Key

What it means

"mcpServers"

Container for all your servers

"my-tools"

A label for this server — you choose the name

"command"

How to run the server (python)

"args"

Full path to your server.py file

⚠️ Use the absolute path to your file. Relative paths will not work.

Save the file and restart Claude Desktop. Your tools will appear in the tools panel automatically.



Step 4 — Test It in Claude

Open Claude Desktop and type:

"Say hello"

Claude will call say_hello and return: Hello, World!


Now try:

"What's the weather in Delhi?"

Claude maps Delhi to the city parameter and calls get_weather automatically.



How Claude Calls Your Tools

Here's the full flow:

  1. You type a message in Claude

  2. Claude reads all registered tool docstrings

  3. It decides which tool matches your intent

  4. It passes your input as function parameters

  5. Your function runs locally on your machine

  6. The return value is sent back to Claude

  7. Claude summarises and displays the result


You never have to call the function yourself — Claude handles the entire invocation.



The 3 "Tools" — Cleared Up

The word tool appears in three different places. Here's what each one means:

Where

What it refers to

"my-tools" in config JSON

The server label — just a name you give the connection

Claude's tools panel (UI)

The list of registered functions Claude discovered

@mcp.tool() in Python

The decorator that registers a function as callable by Claude



What to Build Next

Once you have the basics working, you can extend your MCP server to:

  • Fetch live data from any website or REST API

  • Read and write files on your computer

  • Query a database

  • Automate browser actions

  • Build tools specific to your workflow or product



Go Deeper

This tutorial covers the essentials. If you want to go further:



🎓 Full MCP Course — Codersarts Labs

Take the complete project-based course covering tools with parameters, live API connections, file system control, and building a real production-ready MCP server end to end.




📖 Free Text Tutorial — codersarts.dev

Step-by-step written tutorial with all code snippets, config examples, and common error fixes.





Need Help With Your Project?

Building something with MCP — a portfolio project, SaaS product, AI integration, or internal tool? Our team at Codersarts can help you design, build, and ship it.


  • 1-on-1 mentorship for learning MCP and AI development

  • Project assistance for portfolio and freelance work

  • Full development support for SaaS and production apps




Published by Codersarts — Helping developers build with AI.

Comments

bottom of page