MCP: The USB-C for AI Agents

MCP (Model Context Protocol) is Anthropic’s open standard for connecting AI agents to external tools. Released Nov 2024, hit 1M+ MCP servers by mid-2026.

What is MCP?

Think USB-C for AI:

• One protocol, many tools
• Host (Claude Code, Cursor) speaks MCP
• Servers (Postgres, GitHub, Filesystem) speak MCP
• No custom integration code per host

The architecture

┌────────────┐    MCP     ┌──────────────┐
│   Host     │◄─────────►│  MCP Server  │
│ (Claude)   │   JSON-RPC │ (your tool)  │
└────────────┘            └──────────────┘

Your first MCP server (Python)

pip install mcp

# weather_server.py
from mcp.server import Server
from mcp.types import Tool

app = Server("weather")

@app.tool()
def get_weather(city: str) -> str:
    """Get current weather for a city."""
    return f"Sunny, 72°F in {city}"

app.run()

Connect from Claude Code

Add to ~/.config/claude-code/mcp.json:

{
  "mcpServers": {
    "weather": {
      "command": "python",
      "args": ["weather_server.py"]
    }
  }
}

Now in Claude Code: “What’s the weather in Beijing?” — Claude uses your MCP tool automatically.

Real-world examples

• postgres-mcp: Query your DB from Claude
• github-mcp: Create PRs, read issues from Claude
• playwright-mcp: Browser automation
• github.com/addyosmani/agent-skills: 5 patterns for MCP servers

Key takeaways

• MCP is the standard, not a framework
• One server works across Claude Code, Cursor, Continue.dev
• Tools are auto-discovered by the host