04. Model Context Protocol (MCP) & A2A

class MockMCPServer:
    def __init__(self, name: str):
        self.name = name
        self.users = {
            "user_1": {"name": "Alice", "email": "alice@example.com"},
            "user_2": {"name": "Bob", "email": "bob@example.com"}
        }
 
    def list_tools(self) -> List[Dict]:
        """Discovery: What can this server do?"""
        return [
            {
                "name": "get_user_info",
                "description": "Retrieve user details by ID",
                "input_schema": {
                    "type": "object",
                    "properties": {"user_id": {"type": "string"}},
                    "required": ["user_id"]
                }
            }
        ]
 
    def call_tool(self, tool_name: str, arguments: Dict) -> str:
        """Execute a tool and return results"""
        if tool_name == "get_user_info":
            user = self.users.get(arguments["user_id"])
            return json.dumps(user) if user else "User not found"

class MCPClient:
    def __init__(self, server: MockMCPServer):
        self.server = server
 
    def run(self, query: str):
        # 1. Discovery Phase
        tools = self.server.list_tools()
 
        # 2. Convert to OpenAI format
        openai_tools = [{
            "type": "function",
            "function": {
                "name": t["name"],
                "description": t["description"],
                "parameters": t["input_schema"]
            }
        } for t in tools]
 
        # 3. Let LLM decide and execute
        # ... (agent loop with tool calls)

class ResearchAgent:
    """Agent that exposes itself as a tool"""
 
    def as_tool(self) -> Dict:
        return {
            "name": "ask_researcher",
            "description": "Ask the research agent to find information",
            "input_schema": {
                "type": "object",
                "properties": {
                    "query": {"type": "string"}
                }
            }
        }
 
    def invoke(self, query: str) -> str:
        # This agent does its own research
        return self._search_and_summarize(query)
 
# Manager can now use ResearchAgent as a tool!
manager_tools = [research_agent.as_tool()]

// claude_desktop_config.json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/dir"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token"
      }
    }
  }
}