MCP工具用于开发和优化多智能体AI系统

============================= 一套全面的工具包，用于使用模型上下文协议（MCP），实现AI代理与各种数据源、API和服务之间的无缝集成。无论你是在构建、测试还是部署多代理系统，MCP Kit都能简化工具编排的复杂性，并为开发提供强大的模拟功能。

特色

灵活目标系统

• MCP 服务器：连接现有 MCP 服务器（托管或规范）
• OpenAPI 集成：自动将 REST API 转换为使用 OpenAPI/Swagger 规范的 MCP 工具
• 模拟回答：使用大型语言模型（LLM）或随机生成器生成真实的测试数据
• 复用：将多个目标合并为统一接口

框架适配器

• OpenAI 代理 SDK：原生集成于 OpenAI 代理框架
• LangGraph：LangGraph工作流程的无缝工具集成
• 通用客户端会话：直接MCP协议通信
• 官方MCP服务器：标准MCP服务器包装器

配置驱动架构

MCP Kit Python

MCP tooling for developing and optimizing multi-agent AI systems

A comprehensive toolkit for working with the Model Context Protocol (MCP), providing seamless integration between AI agents and various data sources, APIs, and services. Whether you’re building, testing, or deploying multi-agent systems, MCP Kit simplifies the complexity of tool orchestration and provides powerful mocking capabilities for development.

Features

Flexible Target System

• MCP Servers: Connect to existing MCP servers (hosted or from specifications)
• OpenAPI Integration: Automatically convert REST APIs to MCP tools using OpenAPI/Swagger specs
• Mock Responses: Generate realistic test data using LLM or random generators
• Multiplexing: Combine multiple targets into a unified interface

Framework Adapters

• OpenAI Agents SDK: Native integration with OpenAI’s agent framework
• LangGraph: Seamless tool integration for LangGraph workflows
• Generic Client Sessions: Direct MCP protocol communication
• Official MCP Server: Standard MCP server wrapper

Configuration-Driven Architecture

• YAML/JSON Configuration: Declarative setup for complex workflows
• Factory Pattern: Clean, testable component creation
• Environment Variables: Secure credential management

Advanced Response Generation

• LLM-Powered Mocking: Generate contextually appropriate responses using LLMs
• Random Data Generation: Create test data for development and testing
• Custom Generators: Implement your own response generation logic

---

Quick Start

Installation

uv add mcp-kit

Basic Usage

First you write the Proxy config:

# proxy_config.yaml
""" A mocked REST API target given the OpenAPI spec using LLM-generated responses
"""
target:
  type: mocked
  base_target:
    type: oas
    name: base-oas-server
    spec_url: https://petstore3.swagger.io/api/v3/openapi.json
  tool_response_generator:
    type: llm
    model: openai/gpt-4.1-nano

Don’t forget to setup the LLM API KEY:

# .env
OPENAI_API_KEY="your_openai_key"

Then we can use it as any other MCP:

# main.py
from mcp_kit import ProxyMCP

async def main():
    # Create proxy from configuration
    proxy = ProxyMCP.from_config("proxy_config.yaml")

    # Use with MCP client session adapter
    async with proxy.client_session_adapter() as session:
        tools = await session.list_tools()
        result = await session.call_tool("get_pet", {"pet_id": "777"})
        print(result)

if __name__ == "__main__":
    import asyncio