Quick Start (TL;DR)

The Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI assistants securely query external tools and data sources.

To build a functional MCP server in Python right now, install FastMCP:

pip install "mcp[cli]" fastmcp

Create server.py:

from fastmcp import FastMCP

mcp = FastMCP("SystemUtilities")

@mcp.tool()
def get_disk_usage(path: str = "/") -> str:
    """Return disk usage statistics for the given filesystem path."""
    import shutil
    total, used, free = shutil.disk_usage(path)
    return f"Total: {total // (2**30)}GB | Used: {used // (2**30)}GB | Free: {free // (2**30)}GB"

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

Run and test it immediately in your terminal inspector:

mcp dev server.py

---

1. Connecting Your MCP Server to AI Clients

Once your server script runs locally, you can register it with any MCP-compatible AI client.

Configuration for Claude Code / Claude Desktop

Add the following block to your configuration file (~/.claude/mcp.json or claude_desktop_config.json):

{
  "mcpServers": {
    "system-utils": {
      "command": "python3",
      "args": ["/absolute/path/to/server.py"]
    }
  }
}

Configuration for Antigravity & Cursor

In your workspace configuration or mcp_config.json:

{
  "mcpServers": {
    "system-utils": {
      "type": "stdio",
      "command": "python3",
      "args": ["/absolute/path/to/server.py"]
    }
  }
}

---

2. FastMCP vs Standard Low-Level SDK Comparison

Feature	FastMCP (Recommended)	Official Low-Level MCP SDK
Setup Time	Under 2 minutes	15–20 minutes boilerplate
Type Hinting & Schema Generation	Automatic from Python docstrings	Manual JSON Schema definitions
Transport Support	STDIO and SSE built-in	Separate transport wrappers required
Best For	Rapid custom tools & internal APIs	Enterprise custom transport protocols

---

3. Best Practices for Production MCP Tools

1. Write Descriptive Docstrings: AI agents inspect function docstrings to decide when and how to call a tool. Ambiguous docstrings cause hallucinated argument values.
2. Handle Exceptions Gracefully: Return formatted error strings rather than crashing the process, allowing the agent to self-heal.
3. Keep Tool Payloads Concise: Truncate overly long JSON responses to prevent exhausting the agent’s context window.

---

Summary Checklist

• Install fastmcp for zero-boilerplate tool definitions.
• Expose functions using the @mcp.tool() decorator with clean docstrings.
• Test with mcp dev server.py before linking to your main agent.
• Add absolute file paths in your client config (mcp.json).