Introduction to the Model Context Protocol (MCP) for AI Applications

“MCP is like the USB-C of AI.” – Abdul Wahab, AI Developer

The Model Context Protocol (MCP) is an emerging standard developed by Anthropic that allows developers to connect tools, data, and prompts to AI models in a seamless, reusable way. Whether you’re building a custom LLM app or want to integrate your AI assistant with Slack, Google Drive, or Gmail, MCP offers a universal interface to do so.

In this guide, we’ll break down:

• What MCP is and why it matters
• How MCP works under the hood
• The difference between MCP clients and servers
• How to build your own MCP server using Python
• How to connect it to Claude Desktop

What Is MCP and Why Should You Care?

The Model Context Protocol is an open standard that connects AI applications to tools, context, and resources.

Anthropic compares MCP to USB-C: just like USB-C standardizes how devices connect, MCP standardizes how LLMs access external functionality.

Key Benefits:

• Custom Integrations: Easily connect services like Slack, Google Drive, or databases to your AI apps.
• Portable Toolsets: Build your own AI tools (e.g., prompt templates, utilities) once and use them across different environments like VS Code or Cursor.
• Open Ecosystem: Your app can connect with 3rd-party MCP servers to instantly gain functionality.

How MCP Works: Under the Hood

At its core, MCP uses a client-server architecture:

• MCP Client: Built into the AI app (e.g., Claude Desktop) and sends requests like:
• What tools are available?
• What resources or prompts do you offer?
• MCP Server: An independent system that listens and responds with:
• Tools (functions like sending emails or accessing APIs)
• Resources (files, static data, databases)
• Prompts (templated text instructions)

Think of it like a customer (client) ordering coffee from a barista (server): the client asks for something, the server delivers.

MCP Client Responsibilities

Clients are integrated into AI apps and manage:

• Discovering capabilities of connected servers
• Fetching resources like files or datasets
• Executing tools on behalf of the model
• Delivering prompts to the LLM

💡 Good news: you rarely need to build an MCP client from scratch. Most AI tools (like Claude Desktop) already have one built-in.

Building an MCP Server with Python

Let’s get hands-on and build a custom MCP server using Anthropic’s official Python SDK.

Use Case: Ava, Your Admin Assistant

We'll reuse components from a previous project: Ava, an AI assistant that can:

• Draft Gmail messages
• Access contact directories
• Use templated email prompts

Step 1: Install UV (Universal Environment Manager)

UV is a fast Python package manager built in Rust. It also helps manage your environment with a single command.

Install UV:

curl -Ls https://astral.sh/uv/install.sh | sh

Step 2: Set Up Your MCP Server

Install the SDK, then create the server:

from anthropic_mcp import FastMCP

mcp = FastMCP(name="AVA")

Step 3: Add MCP Components

Add Prompts

@mcp.prompt
def ava_prompt(user_name: str, user_title: str) -> str:
    return f"You are Ava, a virtual assistant for {user_name}, a {user_title}. Help with administrative tasks."

Add Resources

@mcp.resource(uri="email-example/intro")
def intro_email_template() -> str:
    with open("email_examples/three-way-intro.md") as file:
        return file.read()

Add Tools

@mcp.tool
def write_email_draft(subject: str, to: str, body: str) -> str:
    # Connect to Gmail API and create draft
    ...

Step 4: Run the MCP Server

if __name__ == "__main__":
    mcp.run(transport="stdio")  # or use HTTP with SSE

Run the server in dev mode:

uv run mcpd dev MCP_server_example.py

Connecting MCP Server to Claude Desktop

Once your server is running, connect it to Claude Desktop for live testing.

Steps:

1. Open Claude > Settings > Developer
2. Click Edit Config
3. Paste in the MCP server schema with the full path to your UV environment
4. Restart Claude

Now you’ll see:

• Prompts from Ava
• Email draft tools
• Resources like contact directories

You can trigger tools directly from Claude's UI and see results like draft Gmail messages.

Recap: Why MCP is a Game-Changer

✅ Standardized integration with external tools

✅ Reusable prompt and tool sets across apps

✅ Expandable ecosystem with open-source MCP servers

✅ Easy to build your own server using Python and Anthropic SDK

Final Demo

Once connected:

• Claude can request tools like write_email_draft
• It uses prompts and resources from your server
• You can view drafts created in Gmail
• Bonus: You’ll need to explicitly specify which resource or prompt to use in your input