MCP Server Demo Kat

🚀 MCP Server Demo Guide

This guide briefly explains how to create a simple Model Context Protocol (MCP) server and test it using the MCP Inspector and Claude Desktop. The content is based on the README of modelcontextprotocol/python-sdk, with added clarifications.

🚀 Quick Start

Prerequisites

• uv - An extremely fast Python package and project manager written in Rust.

Steps to Create an MCP Server

1. Environment Setup

uv init mcp-server-demo
cd mcp-server-demo

# Specify Python version
echo "3.12" > .python-version

# Create a new virtual environment before installing Python
uv venv 

# Activate the virtual environment
source .venv/bin/activate.fish

# Verify the Python environment (check if the correct Python environment is used)
which python
path/to/mcp-server-demo/.venv/bin/python

# Install MCP dependencies (add to pyproject.toml)
uv add "mcp[cli]"

# Verify that MCP dependencies are installed correctly
uv tree

Resolved 28 packages in 3ms
mcp-server-demo v0.1.0
└── mcp[cli] v1.6.0
    ├── ...

2. Create a Simple MCP Server

# server.py
from mcp.server.fastmcp import FastMCP

# Create an MCP server
mcp = FastMCP("Demo")

# Add an addition tool
@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two numbers"""
    return a + b


# Add a dynamic greeting resource
@mcp.resource("greeting://{name}")
def get_greeting(name: str) -> str:
    """Get a personalized greeting"""
    return f"Hello, {name}!"

3. Test with the MCP Inspector

Start the MCP Inspector

uv run mcp dev server.py

Starting the MCP Inspector...
⚙️  Proxy server listening on port 6277

Test the add Tool

1. Open the MCP Inspector.
2. Click "Connect".
3. Click the "Tools" tab.
4. Click "List Tools" within the "Tools" tab.

Test the get_greeting Resource

1. Open the MCP Inspector.
2. Click "Connect".
3. Click the "Resources" tab.
4. Click "List Templates".
5. Find "get_greeting".
6. Enter a name (e.g., "John").
7. Click "Read Resources".

Example response:

{
  "contents": [
    {
      "uri": "greeting://John",
      "mimeType": "text/plain",
      "text": "Hello, John!"
    }
  ]
}

4. Test with Claude Desktop

Install the MCP Server in Claude Desktop

uv run mcp install server.py

[/dd/MM/YY HH:MM:SS] INFO     Added server 'Demo' to Claude config            claude.py:129
                     INFO     Successfully installed Demo in Claude app          cli.py:467

Verify the Configuration

# For macOS/Linux
cat ~/Library/Application\ Support/Claude/claude_desktop_config.json

Example configuration:

{
  "servers": [
    {
      "name": "Demo",
      "url": "http://localhost:5000"
    }
  ]
}

Use the MCP Server

1. Start your MCP server (e.g., run python server.py).
2. Open Claude Desktop.
3. Enter questions in Claude, such as:
   • "What is 5 plus 3?"
   • "今天是几月？"

Claude Desktop will use the tools provided by the MCP server to answer these questions.

❓ Frequently Asked Questions

How to Add Custom Tools to the MCP Server?

• Define new tools in server.py using the @mcp.tool() decorator.
• Example:

@mcp.tool()
def greet(name: str) -> str:
    """Greet someone"""
    return f"Hello, {name}!"

How to Choose Between Multiple MCP Servers?

• Specify the priority order in the configuration file.
• Example:

{
  "servers": [
    {
      "name": "Primary",
      "url": "http://primary-server:5000"
    },
    {
      "name": "Secondary",
      "url": "http://secondary-server:5000"
    }
  ]
}

📚 References

• Model Context Protocol > Building MCP with LLMs
• Model Context Protocol > Debugging
• Model Context Protocol > Inspector