MCP AI Infra Real Time Agent

🚀 Project Overview

This project focuses on the MCP protocol and its integration with AI and containerization, providing a way to develop and integrate functions into multiple host systems.

🚀 Quick Start

Part 1: MCP Protocol and General AI Interface

• Similar to USB - C, the MCP server acts like an external device that can connect to AI (e.g., Claude Desktop) or cloud applications. We can develop functions once and integrate them into multiple host systems.
• The MCP client is embedded inside AI hosts (such as Claude Desktop, Cursor, etc.) and is responsible for one - to - one interaction with the MCP server. The MCP client can call tools, query resources, and process prompts; while the MCP server exposes tools (model control: retrieval, database update, sending), resources (application control: database records, API), and prompts (user control: documents).

Part 2: MCP and Containerization

• Use Udemy to create a project, set up a virtual environment, install dependencies (such as MCP CLI), and use Cursor to index the official MCP documentation.
• Update the project to comply with Cursor rules.

💻 Usage Examples

1. Implement a Simple MCP Server @server.py

# server.py 实现一个简单的MCP服务器，使用MCP Python SDK，并暴露一个名为terminal的工具，允许用户运行终端命令。

from mcp.mcp import MCPClient, MCPServer
import subprocess

class TerminalTool:
    def __init__(self):
        pass
    
    async def run_command(self, command):
        # 使用subprocess执行终端命令并捕获输出
        process = await asyncio.create_subprocess_shell(
            command,
            stdout=asyncio.PIPE,
            stderr=asyncio.PIPE
        )
        stdout, stderr = await process.communicate()
        result = {
            'command': command,
            'stdout': stdout.decode('utf-8'),
            'stderr': stderr.decode('utf-8'),
            'exit_code': process.returncode
        }
        return result

if __name__ == "__main__":
    # 初始化MCP服务器
    server = MCPServer()
    # 注册terminal工具到服务器
    server.register_tool('terminal', TerminalTool())
    
    print("启动MCP服务器...")
    server.start()

2. Expose Resources in the MCP Server

# server.py 暴露位于桌面目录下的mcpreadme.md文件作为资源，使用MCP Python SDK。

from mcp.mcp import MCPClient, MCPServer
import os

class ReadmeResource:
    def __init__(self):
        self.resource_path = os.path.join(os.path.expanduser('~'), 'Desktop', 'mcpreadme.md')
    
    async def get_readme(self):
        try:
            with open(self.resource_path, 'r') as file:
                content = await file.read()
            return {'content': content}
        except FileNotFoundError:
            return {'error': f'文件 {self.resource_path} 未找到'}
        except IOError:
            return {'error': '读取文件时发生错误'}

if __name__ == "__main__":
    # 初始化MCP服务器
    server = MCPServer()
    # 注册readme资源到服务器
    server.register_resource('mcp_readme', ReadmeResource())
    
    print("启动MCP服务器...")
    server.start()

📚 Documentation

Appendix

• Avoid Reinventing the Wheel: Use existing tools and libraries to simplify the development process.
• Debugging and Testing: Use Anthropic's MCP Inspector for interactive testing and debugging without full deployment.

References

• MCP Crash Course: Master the Model Context Protocol in One Day