LangChain使用mcp-openai和通义千问

1.安装依赖包

pip install langchain
pip install langchain-core
pip install langgraph
pip install langchain_mcp_tools
pip install websockets
# 使用openai大模型需要
pip install langchain-openai
# 使用通义千问大模型需要
pip install --upgrade --quiet langchain-community dashscope

2.安装npx、uvx命令

本机安装node.js和python，npx和uvx用于在本地安装mcp工具包

npm install -g npx
pip install uv

3.目录结构

使用pycharm开发工具

4.环境变量

OPENAI_API_KEY="填写自己的openai申请的api_key,我是使用下面的第三方代理申请的"

OPENAI_BASE_URL="https://api.closeai-proxy.xyz/v1"

DASHSCOPE_API_KEY="填写自己申请的通义千问api_key"

5.python代码

import asyncio
import logging
import os
import sys

# Third-party imports
try:
    from dotenv import load_dotenv, find_dotenv

    from langchain_core.messages import HumanMessage
    from langchain_openai import ChatOpenAI
    from langchain_community.chat_models import ChatTongyi
    from langgraph.prebuilt import create_react_agent
except ImportError as e:
    print(f'\nError: Required package not found: {e}')
    print('Please ensure all required packages are installed\n')
    sys.exit(1)

# Local application imports
from langchain_mcp_tools import convert_mcp_to_langchain_tools


# A very simple logger
def init_logger() -> logging.Logger:
    logging.basicConfig(
        level=logging.INFO,  # logging.DEBUG,
        format='\x1b[90m[%(levelname)s]\x1b[0m %(message)s'
    )
    return logging.getLogger()


async def run() -> None:
    # Be sure to set ANTHROPIC_API_KEY and/or OPENAI_API_KEY as needed
    global cleanup
    _ = load_dotenv(find_dotenv())

    try:
        mcp_configs = {
            'filesystem': {
                'command': 'npx',
                'args': [
                    '-y',
                    '@modelcontextprotocol/server-filesystem',
                    '.'  # path to a directory to allow access to
                ]
            },
            'fetch': {
                'command': 'uvx',
                'args': [
                    'mcp-server-fetch'
                ]
            },
            'weather': {
                'command': 'npx',
                'args': [
                    '-y',
                    '@h1deya/mcp-server-weather'
                ]
            },
            "amap-maps": {
              "command": "npx",
              "args": [
                "-y",
                "@amap/amap-maps-mcp-server"
              ],
              "env": {
                "AMAP_MAPS_API_KEY": "8d20d787d32595a9542983ced6fce5e1"
              }
            },
            "tavily-mcp": {
              "command": "npx",
              "args": [
                "-y",
                "tavily-mcp"
              ],
              "env": {
                "TAVILY_API_KEY": "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"
              },
            },
            "mysql": {
                "type": "stdio",
                "command": "uvx",
                "args": [
                    "--from",
                    "mysql-mcp-server",
                    "mysql_mcp_server"
                ],
                "env": {
                    "MYSQL_HOST": "172.16.0.114",
                    "MYSQL_PORT": "3308",
                    "MYSQL_USER": "ds",
                    "MYSQL_PASSWORD": "ds123",
                    "MYSQL_DATABASE": "settlebi"
                }
            },
        }
        tools, cleanup = await convert_mcp_to_langchain_tools(
            mcp_configs,
            init_logger()
        )
        # llm = ChatOpenAI(model="gpt-4o", temperature=0)
        llm = ChatTongyi(temperature=0)
        agent = create_react_agent(
            llm,
            tools
        )

        while True:
            query = input("请输入您的问题 (输入 '退出'、'exit' 或 'quit' 来结束): ")
            if query.lower() in ['退出', 'exit', 'quit']:
                print("正在退出...")
                break

            if not query.strip():
                print("问题不能为空，请重新输入。")
                continue

            print('问题：', query)
            messages = [HumanMessage(content=query)]
            try:
                result = await agent.ainvoke({'messages': messages})
                # the last message should be an AIMessage
                response = result['messages'][-1].content
                print('大模型输出结果开始')
                print(response)
                print('大模型输出结果结束')
            except Exception as e:
                print(f"处理请求时发生错误: {e}")
            print('--------------------')  # 分隔每次问答
    finally:
        if cleanup:
            await cleanup()

        if sys.platform == "win32":
            await asyncio.sleep(0.25)

        print('------- 清理完成，程序结束 -------')


def main() -> None:
    asyncio.run(run())


if __name__ == '__main__':
    main()