MCP简介

MCP（Model Context Protocol） 是由 Anthropic 公司提出并开源的一个标准协议，旨在解决大型语言模型（LLMs）与外部数据源和工具集成的难题。它为AI模型提供了一个通用接口，使得这些模型能够安全、高效地访问实时数据，并执行操作，从而不再局限于训练时所包含的静态知识。

MCP可以被视为AI领域的“USB-C接口”，它通过标准化的方式让AI助手能够连接到任何数据源或服务，就像USB-C统一了电子设备与各种配件的连接方式一样。该协议基于客户端-服务器架构，支持Python、TypeScript等多种编程语言的SDK，方便开发者快速构建服务。

根据博主自己的理解，MCP和Langchain的tools非常的相似。但MCP更具有通用性。Langchain tools只能归Langchain框架所用，但MCP是完全通用的，并不是某一些框架的专属。除此之外MCP还支持部署为服务，提供远程调用的能力。易用性和可复用能力更胜一筹。

MCP协议

在 MCP（Model Context Protocol）的上下文中，Stdio 和 SSE 是两种不同的通信方式，它们各自有其特点和适用场景。以下是这两种通信方式的特点以及对比：

Stdio (标准输入输出)

特点：

• 本地进程间通信：Stdio 通常用于同一台机器上的进程间通信。
• 低延迟：由于数据不需要通过网络传输，因此延迟非常低。
• 安全性高：因为通信仅限于本地进程之间，所以减少了外部攻击的风险。
• 简单直接：无需复杂的配置，适合快速开发和测试。
• 强耦合性：客户端和服务端必须在同一主机上运行，形成一对一的关系。

适用场景：

• 客户端和服务器在同一台机器上运行。
• 需要本地访问文件系统或设备的工具。
• 对延迟敏感的应用。
• 开发和测试阶段。

SSE (Server-Sent Events)

特点：

• 基于 HTTP 的单向通信：SSE 允许服务器主动向客户端推送数据，但客户端不能通过相同的连接发送数据给服务器。
• 轻量级实现：易于实现且适用于需要实时更新的应用场景，如新闻推送、股票行情等。
• 自动重连：如果连接断开，客户端会自动尝试重新连接，确保数据流的连续性。
• 支持多客户端：单个服务器实例可以同时服务于多个客户端，具有良好的扩展性。
• 跨网络通信：适用于客户端和服务器不在同一物理位置的情况。

适用场景：

• 分布式系统或远程服务调用。
• 支持多客户端连接的环境。
• 需要远程访问的服务。
• 生产环境和企业级应用。

对比

MCP服务端

接下来我们分别编写两种类型的MCP服务端。

安装依赖

uv add mcp mcp[cli]
# 该依赖是服务所需，对于MCP框架不是必须的
uv add requests

编写MCP服务端

STDIO协议服务端：demo.py。编写两个工具，分别为生成问候语和生成HDP大数据组件的安装路径。例子较为简单，代码如下所示：

from mcp.server.fastmcp import FastMCP
import os

mcp = FastMCP()

@mcp.tool()
def say_hello(name: str) -> str:
    """生成问候语"""
    return f"  你好 {name}! (Hello {name}!)"
    

@mcp.tool()
def generate_hdp_install_path(component: str) -> str:
    """获取组件在HDP的安装路径"""
    return os.path.join('/usr/hdp/3.0.1.0-187/', component)


if __name__ == "__main__":
    mcp.run(transport='stdio')

SSE协议服务端：weather.py。这个例子将城市名称转化为城市代码，然后调用天气Rest服务返回天气预报。天气相关部分的代码来源于网络。

from mcp.server.fastmcp import FastMCP

import requests, json


mcp = FastMCP()
city_json = '''
{
    "七台河": "101051002",
    "万宁": "101310215",
    "万州天城": "101041200",
    "万州龙宝": "101041300",
    "万盛": "101040600",
    "三亚": "101310201",
    "三明": "101230801",
    "三门峡": "101181701",
    "上海": "101020100",
    "上饶": "101240301",
    "东丽": "101030400",
    "东方": "101310202",
    "东莞": "101281601",
    "东营": "101121201",
    "中卫": "101170501",
    "中山": "101281701",
    "丰台": "101010900",
    "丰都": "101043000",
    "临夏": "101161101",
    "临汾": "101100701",
    "临沂": "101120901",
    "临沧": "101291101",
    "临河": "101080801",
    "临高": "101310203",
    "丹东": "101070601",
    "丽水": "101210801",
    "丽江": "101291401",
    "乌兰浩特": "101081101",
    "乌海": "101080301",
    "乌鲁木齐": "101130101",
    "乐东": "101310221",
    "乐山": "101271401",
    "九江": "101240201",
    "云浮": "101281401",
    "云阳": "101041700",
    "五指山": "101310222",
    "亳州": "101220901",
    "仙桃": "101201601",
    "伊宁": "101131001",
    "伊春": "101050801",
    "佛山": "101280800",
    "佛爷顶": "101011700",
    "佳木斯": "101050401",
    "保亭": "101310214",
    "保定": "101090201",
    "保山": "101290501",
    "信阳": "101180601",
    "儋州": "101310205",
    "克拉玛依": "101130201",
    "八达岭": "101011600",
    "六安": "101221501",
    "六盘水": "101260801",
    "兰州": "101160101",
    "兴义": "101260906",
    "内江": "101271201",
    "凉山": "101271601",
    "凯里": "101260501",
    "包头": "101080201",
    "北京": "101010100",
    "北京城区": "101012200",
    "北海": "101301301",
    "北碚": "101040800",
    "北辰": "101030600",
    "十堰": "101201101",
    "南京": "101190101",
    "南充": "101270501",
    "南宁": "101300101",
    "南川": "101040400",
    "南平": "101230901",
    "南昌": "101240101",
    "南汇": "101020600",
    "南沙岛": "101310220",
    "南通": "101190501",
    "南阳": "101180701",
    "博乐": "101131601",
    "厦门": "101230201",
    "双鸭山": "101051301",
    "台中": "101340401",
    "台北县": "101340101",
    "台州": "101210601",
    "合作": "101161201",
    "合川": "101040300",
    "合肥": "101220101",
    "吉安": "101240601",
    "吉林": "101060201",
    "吉首": "101251501",
    "吐鲁番": "101130501",
    "吕梁": "101101100",
    "吴忠": "101170301",
    "周口": "101181401",
    "呼伦贝尔": "101081000",
    "呼和浩特": "101080101",
    "和田": "101131301",
    "咸宁": "101200701",
    "咸阳": "101110200",
    "哈密": "101131201",
    "哈尔滨": "101050101",
    "唐山": "101090501",
    "商丘": "101181001",
    "商洛": "101110601",
    "喀什": "101130901",
    "嘉兴": "101210301",
    "嘉定": "101020500",
    "嘉峪关": "101161401",
    "四平": "101060401",
    "固原": "101170401",
    "垫江": "101042200",
    "城口": "101041600",
    "塔城": "101131101",
    "塘沽": "101031100",
    "大兴": "101011100",
    "大兴安岭": "101050701",
    "大同": "101100201",
    "大庆": "101050901",
    "大港": "101031200",
    "大理": "101290201",
    "大足": "101042600",
    "大连": "101070201",
    "天水": "101160901",
    "天津": "101030100",
    "天门": "101201501",
    "太原": "101100101",
    "奉节": "101041900",
    "奉贤": "101021000",
    "威海": "101121301",
    "娄底": "101250801",
    "孝感": "101200401",
    "宁德": "101230301",
    "宁河": "101030700",
    "宁波": "101210401",
    "安庆": "101220601",
    "安康": "101110701",
    "安阳": "101180201",
    "安顺": "101260301",
    "定安": "101310209",
    "定西": "101160201",
    "宜宾": "101271101",
    "宜昌": "101200901",
    "宜春": "101240501",
    "宝坻": "101030300",
    "宝山": "101020300",
    "宝鸡": "101110901",
    "宣城": "101221401",
    "宿州": "101220701",
    "宿迁": "101191301",
    "密云": "101011300",
    "密云上甸子": "101011900",
    "屯昌": "101310210",
    "山南": "101140301",
    "岳阳": "101251001",
    "崇左": "101300201",
    "崇明": "101021100",
    "巢湖": "101221601",
    "巫山": "101042000",
    "巫溪": "101041800",
    "巴中": "101270901",
    "巴南": "101040900",
    "常州": "101191101",
    "常德": "101250601",
    "平凉": "101160301",
    "平谷": "101011500",
    "平顶山": "101180501",
    "广元": "101272101",
    "广安": "101270801",
    "广州": "101280101",
    "庆阳": "101160401",
    "库尔勒": "101130601",
    "廊坊": "101090601",
    "延吉": "101060301",
    "延安": "101110300",
    "延庆": "101010800",
    "开县": "101041500",
    "开封": "101180801",
    "张家口": "101090301",
    "张家界": "101251101",
    "张掖": "101160701",
    "彭水": "101043200",
    "徐家汇": "101021200",
    "徐州": "101190801",
    "德宏": "101291501",
    "德州": "101120401",
    "德阳": "101272001",
    "忠县": "101042400",
    "忻州": "101101001",
    "怀化": "101251201",
    "怀柔": "101010500",
    "怒江": "101291201",
    "恩施": "101201001",
    "惠州": "101280301",
    "成都": "101270101",
    "房山": "101011200",
    "扬州": "101190601",
    "承德": "101090402",
    "抚州": "101240401",
    "抚顺": "101070401",
    "拉萨": "101140101",
    "揭阳": "101281901",
    "攀枝花": "101270201",
    "文山": "101290601",
    "文昌": "101310212",
    "斋堂": "101012000",
    "新乡": "101180301",
    "新余": "101241001",
    "无锡": "101190201",
    "日喀则": "101140201",
    "日照": "101121501",
    "昆明": "101290101",
    "昌吉": "101130401",
    "昌平": "101010700",
    "昌江": "101310206",
    "昌都": "101140501",
    "昭通": "101291001",
    "晋中": "101100401",
    "晋城": "101100601",
    "晋江": "101230509",
    "普洱": "101290901",
    "景德镇": "101240801",
    "景洪": "101291601",
    "曲靖": "101290401",
    "朔州": "101100901",
    "朝阳": "101071201",
    "本溪": "101070501",
    "来宾": "101300401",
    "杭州": "101210101",
    "松原": "101060801",
    "松江": "101020900",
    "林芝": "101140401",
    "果洛": "101150501",
    "枣庄": "101121401",
    "柳州": "101300301",
    "株洲": "101250301",
    "桂林": "101300501",
    "梁平": "101042300",
    "梅州": "101280401",
    "梧州": "101300601",
    "楚雄": "101290801",
    "榆林": "101110401",
    "武威": "101160501",
    "武汉": "101200101",
    "武清": "101030200",
    "武都": "101161001",
    "武隆": "101043100",
    "毕节": "101260701",
    "永川": "101040200",
    "永州": "101251401",
    "汉中": "101110801",
    "汉沽": "101030800",
    "汕头": "101280501",
    "汕尾": "101282101",
    "江津": "101040500",
    "江门": "101281101",
    "池州": "101221701",
    "汤河口": "101011800",
    "沈阳": "101070101",
    "沙坪坝": "101043700",
    "沧州": "101090701",
    "河池": "101301201",
    "河源": "101281201",
    "泉州": "101230501",
    "泰安": "101120801",
    "泰州": "101191201",
    "泸州": "101271001",
    "洛阳": "101180901",
    "津南": "101031000",
    "济南": "101120101",
    "济宁": "101120701",
    "济源": "101181801",
    "浦东": "101021300",
    "海东": "101150201",
    "海北": "101150801",
    "海南": "101150401",
    "海口": "101310101",
    "海淀": "101010200",
    "海西": "101150701",
    "涪陵": "101041400",
    "淄博": "101120301",
    "淮北": "101221201",
    "淮南": "101220401",
    "淮安": "101190901",
    "深圳": "101280601",
    "清远": "101281301",
    "渝北": "101040700",
    "温州": "101210701",
    "渭南": "101110501",
    "湖州": "101210201",
    "湘潭": "101250201",
    "湛江": "101281001",
    "滁州": "101221101",
    "滨州": "101121101",
    "漯河": "101181501",
    "漳州": "101230601",
    "潍坊": "101120601",
    "潜江": "101201701",
    "潮州": "101281501",
    "潼南": "101042100",
    "澄迈": "101310204",
    "濮阳": "101181301",
    "烟台": "101120501",
    "焦作": "101181101",
    "牡丹江": "101050301",
    "玉林": "101300901",
    "玉树": "101150601",
    "玉溪": "101290701",
    "珠海": "101280701",
    "琼中": "101310208",
    "琼山": "101310102",
    "琼海": "101310211",
    "璧山": "101042900",
    "甘孜": "101271801",
    "白城": "101060601",
    "白山": "101060901",
    "白沙": "101310207",
    "白银": "101161301",
    "百色": "101301001",
    "益阳": "101250700",
    "盐城": "101190701",
    "盘锦": "101071301",
    "眉山": "101271501",
    "石嘴山": "101170201",
    "石家庄": "101090101",
    "石景山": "101011000",
    "石柱": "101042500",
    "石河子": "101130301",
    "神农架": "101201201",
    "福州": "101230101",
    "秀山": "101043600",
    "秦皇岛": "101091101",
    "綦江": "101043300",
    "红河": "101290301",
    "绍兴": "101210501",
    "绥化": "101050501",
    "绵阳": "101270401",
    "聊城": "101121701",
    "肇庆": "101280901",
    "自贡": "101270301",
    "舟山": "101211101",
    "芜湖": "101220301",
    "苏州": "101190401",
    "茂名": "101282001",
    "荆州": "101200801",
    "荆门": "101201401",
    "荣昌": "101042700",
    "莆田": "101230401",
    "莱芜": "101121601",
    "菏泽": "101121001",
    "萍乡": "101240901",
    "营口": "101070801",
    "葫芦岛": "101071401",
    "蓟县": "101031400",
    "蚌埠": "101220201",
    "衡水": "101090801",
    "衡阳": "101250401",
    "衢州": "101211001",
    "襄樊": "101200201",
    "西宁": "101150101",
    "西安": "101110101",
    "西沙": "101310217",
    "西青": "101030500",
    "许昌": "101180401",
    "贵港": "101300801",
    "贵阳": "101260101",
    "贺州": "101300701",
    "资阳": "101271301",
    "赣州": "101240701",
    "赤峰": "101080601",
    "辽源": "101060701",
    "辽阳": "101071001",
    "达州": "101270601",
    "运城": "101100801",
    "连云港": "101191001",
    "通化": "101060501",
    "通州": "101010600",
    "通辽": "101080501",
    "遂宁": "101270701",
    "遵义": "101260201",
    "邢台": "101090901",
    "那曲": "101140601",
    "邯郸": "101091001",
    "邵阳": "101250901",
    "郑州": "101180101",
    "郴州": "101250501",
    "都匀": "101260401",
    "鄂尔多斯": "101080701",
    "鄂州": "101200301",
    "酉阳": "101043400",
    "酒泉": "101160801",
    "重庆": "101040100",
    "金华": "101210901",
    "金山": "101020700",
    "金昌": "101160601",
    "钦州": "101301101",
    "铁岭": "101071101",
    "铜仁": "101260601",
    "铜川": "101111001",
    "铜梁": "101042800",
    "铜陵": "101221301",
    "银川": "101170101",
    "锡林浩特": "101080901",
    "锦州": "101070701",
    "镇江": "101190301",
    "长寿": "101041000",
    "长春": "101060101",
    "长沙": "101250101",
    "长治": "101100501",
    "门头沟": "101011400",
    "闵行": "101020200",
    "阜新": "101070901",
    "阜阳": "101220801",
    "防城港": "101301401",
    "阳江": "101281801",
    "阳泉": "101100301",
    "阿克苏": "101130801",
    "阿勒泰": "101131401",
    "阿图什": "101131501",
    "阿坝": "101271901",
    "阿拉善左旗": "101081201",
    "阿拉尔": "101130701",
    "阿里": "101140701",
    "陵水": "101310216",
    "随州": "101201301",
    "雅安": "101271701",
    "集宁": "101080401",
    "霞云岭": "101012100",
    "青岛": "101120201",
    "青浦": "101020800",
    "静海": "101030900",
    "鞍山": "101070301",
    "韶关": "101280201",
    "顺义": "101010400",
    "香格里拉": "101291301",
    "马鞍山": "101220501",
    "驻马店": "101181601",
    "高雄": "101340201",
    "鸡西": "101051101",
    "鹤壁": "101181201",
    "鹤岗": "101051201",
    "鹰潭": "101241101",
    "黄冈": "101200501",
    "黄南": "101150301",
    "黄山": "101221001",
    "黄石": "101200601",
    "黑河": "101050601",
    "黔江": "101041100",
    "黔阳": "101251301",
    "齐齐哈尔": "101050201",
    "龙岩": "101230701"
}
'''


@mcp.tool()
def get_weather(city):
    city_dict = json.loads(city_json)
    city_code = city_dict.get(city)
    url = 'http://t.weather.sojson.com/api/weather/city/'
    response = requests.get(url + city_code)
    return response.json()


if __name__ == "__main__":
    mcp.run(transport='sse')

执行如下命令，启动SSE服务端：

uv run weather.py

成功启动后能看到类似如下日志输出：

INFO:     Started server process [2295781]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)

日志告知SSE服务的访问地址为http://0.0.0.0:8000。后面调用的时候需要。

注意：如果需要指定端口号，可在创建FastMCP示例的时候指定，例如mcp = FastMCP(port="10080")。

Langchain调用端

安装依赖

创建另一个Python项目，安装依赖。

uv add langchain langchain-ollama langchain-mcp-adapters langgraph

编写调用端代码

环境要求：需要使用Ollama较新版本，例如0.5.7以上。使用较低版本（例如0.3.13）可能无法支持funtion call，导致下面例子无法调用MCP。

main.py内容如下：

from langgraph.prebuilt import create_react_agent
from langchain_ollama import ChatOllama
from langchain_mcp_adapters.client import MultiServerMCPClient

import asyncio

async def main():
    BASE_URL="http://ollama_ip:port"
    llm = ChatOllama(base_url=BASE_URL, model="qwen2.5:7b", temperature=0.2)
    async with MultiServerMCPClient({
        "stdio_server": {
            "command": "uv",
            "args": ["run", "/home/paul/Documents/mcp-servers/demo.py"],
            "transport": "stdio"
        },
        "weather_server": {
            "url": "http://mcp_server_ip:8000/sse",
            "transport": "sse"
        }
    }) as client:
        agent = create_react_agent(llm, client.get_tools())
        hello_response = await agent.ainvoke({"messages": "为Paul打招呼"})
        path_response = await agent.ainvoke({"messages": "get Flink installation path in hdp"})
        weather_response = await agent.ainvoke({"messages": "武汉天气怎样？"})
        print("======================================")
        print(hello_response)
        print(path_response)
        print(weather_response)


if __name__ == "__main__":
    asyncio.run(main())

需要注意的是，上面的写法对于langchain-mcp-adapters==v0.0.9版本有效。如果使用langchain-mcp-adapters==v0.1.0以上版本，需要修改为：

from langgraph.prebuilt import create_react_agent
from langchain_ollama import ChatOllama
from langchain_mcp_adapters.client import MultiServerMCPClient

import asyncio

async def main():
    BASE_URL="http://ollama_ip:port"
    llm = ChatOllama(base_url=BASE_URL, model="qwen2.5:7b", temperature=0.2)
    client = MultiServerMCPClient({
        "stdio_server": {
            "command": "uv",
            "args": ["run", "/home/paul/Documents/mcp-servers/demo.py"],
            "transport": "stdio"
        },
        "weather_server": {
            "url": "http://mcp_server_ip:8000/sse",
            "transport": "sse"
        }
    })
    agent = create_react_agent(llm, await client.get_tools())
    hello_response = await agent.ainvoke({"messages": "为Paul打招呼"})
    path_response = await agent.ainvoke({"messages": "get Flink installation path in hdp"})
    weather_response = await agent.ainvoke({"messages": "武汉天气怎样？"})
    print("======================================")
    print(hello_response)
    print(path_response)
    print(weather_response)


if __name__ == "__main__":
    asyncio.run(main())

后面的示例均使用v0.0.9版本。如果使用v0.1.0及以上版本，也要如此修改，不再赘述。

运行调用端

uv run main.py

输出如下所示：

{'messages': [HumanMessage(content='为Paul打招呼', additional_kwargs={}, response_metadata={}, id='8fc409b6-45da-44ef-9e6b-b84ebafa8cd3'), AIMessage(content='', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:42.667135103Z', 'done': True, 'done_reason': 'stop', 'total_duration': 4654188369, 'load_duration': 3876471088, 'prompt_eval_count': 293, 'prompt_eval_duration': 168000000, 'eval_count': 20, 'eval_duration': 180000000, 'model_name': 'qwen2.5:7b'}, id='run-f1dc5578-c1c1-4b24-b5b4-a57c687e7a43-0', tool_calls=[{'name': 'say_hello', 'args': {'name': 'Paul'}, 'id': '5a02b05e-513c-4cf7-befd-d401072d50b7', 'type': 'tool_call'}], usage_metadata={'input_tokens': 293, 'output_tokens': 20, 'total_tokens': 313}), ToolMessage(content='  你好 Paul! (Hello Paul!)', name='say_hello', id='a22ec4e2-3ac5-4de2-8ed0-bbe3b6b995c5', tool_call_id='5a02b05e-513c-4cf7-befd-d401072d50b7'), AIMessage(content='你好 Paul! (Hello Paul!)', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:42.841418827Z', 'done': True, 'done_reason': 'stop', 'total_duration': 141787719, 'load_duration': 15252931, 'prompt_eval_count': 338, 'prompt_eval_duration': 17000000, 'eval_count': 8, 'eval_duration': 76000000, 'model_name': 'qwen2.5:7b'}, id='run-d7579ce5-d961-4ec4-b0ca-dbc53b5c53e8-0', usage_metadata={'input_tokens': 338, 'output_tokens': 8, 'total_tokens': 346})]}
{'messages': [HumanMessage(content='get Flink installation path in hdp', additional_kwargs={}, response_metadata={}, id='3eef711b-a47e-4807-bd5f-75c934e239a1'), AIMessage(content='', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:43.130010048Z', 'done': True, 'done_reason': 'stop', 'total_duration': 258172946, 'load_duration': 15869845, 'prompt_eval_count': 298, 'prompt_eval_duration': 11000000, 'eval_count': 24, 'eval_duration': 222000000, 'model_name': 'qwen2.5:7b'}, id='run-9ba1cf72-15a5-4803-bdeb-6ce3a2828170-0', tool_calls=[{'name'10: 'generate_hdp_install_path', 'args': {'component': 'flink'}, 'id': '0ab08f1c-fc43-4615-8f71-baa3fb7a26be', 'type': 'tool_call'}], usage_metadata={'input_tokens': 298, 'output_tokens': 24, 'total_tokens': 322}), ToolMessage(content='/usr/hdp/3.0.1.0-187/flink', name='generate_hdp_install_path', id='c01330d1-ad5a-4199-a721-48e3dfc43cbc', tool_call_id='0ab08f1c-fc43-4615-8f71-baa3fb7a26be'), AIMessage(content='The installation path for Flink in your HDP is `/usr/hdp/3.0.1.0-187/flink`.', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:43.513232Z', 'done': True, 'done_reason': 'stop', 'total_duration': 353325750, 'load_duration': 14993601, 'prompt_eval_count': 356, 'prompt_eval_duration': 16000000, 'eval_count': 31, 'eval_duration': 289000000, 'model_name': 'qwen2.5:7b'}, id='run-326e9b3e-85f9-4ae7-ac05-30aa40298974-0', usage_metadata={'input_tokens': 356, 'output_tokens': 31, 'total_tokens': 387})]}
{'messages': [HumanMessage(content='武汉天气怎样？', additional_kwargs={}, response_metadata={}, id='61597a94-9122-4494-856a-91fe909ce8c6'), AIMessage(content='', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:43.762996227Z', 'done': True, 'done_reason': 'stop', 'total_duration': 219408503, 'load_duration': 15236008, 'prompt_eval_count': 294, 'prompt_eval_duration': 11000000, 'eval_count': 20, 'eval_duration': 184000000, 'model_name': 'qwen2.5:7b'}, id='run-daca685a-29a6-4a5f-8bf4-71795eb621da-0', tool_calls=[{'name': 'get_weather',

... (太长省略)

), AIMessage(content='根据提供的信息，接下来一周的天气情况如下：\n\n- 4月28日（星期六）：最高温度26°C，最低温度19°C。风向为西南风，风力等级为二级。天气状况为多云，注意防寒保暖。\n- 4月29日（星期天）：最高温度27°C，最低温度20°C。风向为东南风，风力等级为三级。天气状况为晴朗，气温适宜，适合户外活动。\n- 4月30日（星期一）：最高温度28°C，最低温度21°C。风向为东北风，风力等级为二级。天气状况为多云，注意防晒。\n- 5月1日（星期二）：最高温度29°C，最低温度22°C。风向为东南风，风力等级为三级。天气状况为晴朗，气温较高，请适当减少衣物。\n- 5月2日（星期三）：最高温度30°C，最低温度23°C。风向为东北风，风力等级为二级。天气状况为多云，注意防晒。\n- 5月3日（星期四）：最高温度31°C，最低温度24°C。风向为东南风，风力等级为三级。天气状况为晴朗，气温较高，请适当减少衣物。\n- 5月4日（星期五）：最高温度32°C，最低温度25°C。风向为东北风，风力等级为二级。天气状况为多云，注意防晒。\n- 5月5日（星期六）：最高温度31°C，最低温度26°C。风向为东南风，风力等级为三级。天气状况为晴朗，气温较高，请适当减少衣物。\n- 5月6日（星期天）：最高温度32°C，最低温度27°C。风向为东北风，风力等级为二级。天气状况为多云，注意防晒。\n\n建议：\n1. 多数日子的天气以晴朗或多云为主，请根据天气情况合理安排户外活动。\n2. 由于气温较高，请注意防暑降温，并适当减少衣物。\n3. 注意防晒措施，避免长时间暴露在阳光下。\n4. 风力较大时请注意安全，尽量避免在强风条件下进行户外活动。', additional_kwargs={}, response_metadata={'model': 'qwen2.5:7b', 'created_at': '2025-04-28T02:06:48.946271634Z', 'done': True, 'done_reason': 'stop', 'total_duration': 5085808347, 'load_duration': 16067136, 'prompt_eval_count': 2048, 'prompt_eval_duration': 258000000, 'eval_count': 504, 'eval_duration': 4797000000, 'model_name': 'qwen2.5:7b'}, id='run-9cbbb3ba-487d-49b9-a1e8-e7b329aadb16-0', usage_metadata={'input_tokens': 2048, 'output_tokens': 504, 'total_tokens': 2552})]}

三个MCP服务均调用成功。

获取Prompt

Prompts 是MCP服务器提供的结构化模板，旨在标准化与语言模型的交互。它们允许定义可重复使用的消息序列和工作流程，并且可以通过参数来定制每次交互的具体内容。用于提供一致的提示工程框架，确保 AI 模型以可预测的方式响应。允许用户或客户端通过参数（如用户名、风格等）动态生成提示，从而引导 AI 模型生成特定输出或执行特定行为。适用于需要引导 AI 行为或生成特定输出的场景，例如生成个性化问候、格式化查询等。

接下来我们编写一个例子，在Langchain中调用MCP服务获取指定的prompt。

继续上面的例子，在demo.py中增加：

@mcp.prompt()
def code_review_prompt(code: str) -> str:
    return f"指出以下代码中的问题：\n\n{code}"

修改Langchain调用端main.py：

from langchain_ollama import ChatOllama
from langchain_mcp_adapters.client import MultiServerMCPClient

import asyncio

async def main():
    BASE_URL="http://ollama_ip:port"
    llm = ChatOllama(base_url=BASE_URL, model="qwen2.5:7b", temperature=0.2)
    async with MultiServerMCPClient({
        "stdio_server": {
            "command": "uv",
            "args": ["run", "/home/paul/Documents/mcp-servers/demo.py"],
            "transport": "stdio"
        }
    }) as client:
        print(await client.get_prompt("stdio_server", "code_review_prompt", {"code": "System.out,println(someStr)"}))

if __name__ == "__main__":
    asyncio.run(main())

get_prompt方法定义为：

get_prompt(
    server_name: str,
    prompt_name: str,
    arguments: Optional[dict[str, Any]],
) -> list[HumanMessage | AIMessage]

其中：

• server_name: MCP服务器名称。对应创建MultiServerMCPClient指定的key。
• prompt_name: Prompt名称。对应@mcp.prompt()装饰的方法名。
• arguments: 参数列表。对应@mcp.prompt()装饰的方法的参数列表。

执行调用端之后得到如下输出：

[HumanMessage(content='指出以下代码中的问题：\n\nSystem.out,println(someStr)', additional_kwargs={}, response_metadata={})]

参考链接：https://github.langchain.ac.cn/langgraph/reference/mcp/#langchain_mcp_adapters.client.MultiServerMCPClient.get_prompt

获取Resource

Resources 代表MCP server想要提供给clients的任何类型的数据。这可以是文件、数据库记录、API 响应等任何形式的数据。它们通过唯一的URI标识，可以是文本（UTF-8 编码）或者二进制数据（Base64 编码）。用于提供上下文数据给 AI 模型，如日志文件、配置文件、实时 API 数据等。支持动态数据更新，服务器可以通过通知机制告知客户端资源的变化。适合需要结构化、可重复访问的数据场景，例如文件内容、数据库查询结果等。

接下来我们编写一个例子，在Langchain中调用MCP服务获取指定的resource。

继续上面的例子，在demo.py中增加：

@mcp.resource("config://app_settings")
def get_app_config() -> dict:
    return {"theme": "dark", "language": "zh-CN"}

修改Langchain调用端main.py：

from langchain_ollama import ChatOllama
from langchain_mcp_adapters.client import MultiServerMCPClient

import asyncio

async def main():
    BASE_URL="http://ollama_ip:port"
    llm = ChatOllama(base_url=BASE_URL, model="qwen2.5:7b", temperature=0.2)
    async with MultiServerMCPClient({
        "stdio_server": {
            "command": "uv",
            "args": ["run", "/home/paul/Documents/mcp-servers/demo.py"],
            "transport": "stdio"
        }
    }) as client:
        resource = await client.get_resources("stdio_server", "config://app_settings")
        print(resource[0].as_string())


if __name__ == "__main__":
    asyncio.run(main())

get_resource方法定义为：

get_resources(
    server_name: str, uris: str | list[str] | None = None
) -> list[Blob]

其中：

• server_name: MCP服务器名称。对应创建MultiServerMCPClient指定的key。
• uris: Resource对应的URI。为@mcp.resource装饰器中的值。

执行调用端之后得到如下输出：

{"theme": "dark", "language": "zh-CN"}

参考链接：https://github.langchain.ac.cn/langgraph/reference/mcp/#langchain_mcp_adapters.client.MultiServerMCPClient.get_resources

传递环境变量

MCP服务使用时很可能需要环境信息。例如使用操作MySQL的MCP Server，需要配置MySQL的连接信息。这种信息传递给MCP Server最好的方式就是通过环境变量传递。

注意：只有STDIO模式的MCP Server支持通过环境变量方式传递配置信息。SSE方式不能通过此方式，只能在启动时候指定。

下面例子中我们通过os.getenv()方法读取环境变量，如下所示：

from mcp.server.fastmcp import FastMCP
import os
import logging


logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
mcp = FastMCP()


def get_config():
    config = {
        "mysql_host": os.getenv("MYSQL_HOST", "127.0.0.1"),
        "db_name": os.getenv("DB_NAME", "mydb"),
        "mysql_user": os.getenv("MYSQL_USER", "paul"),
        "mysql_password": os.getenv("MYSQL_PASSWORD", "123456"),
    }
    logger.info(config)
    # 其他操作...
    return config


@mcp.tool()
def list_tables() -> list:
    config = get_config()
    # 其他操作...

在LangChain配置MultiServerMCPClient时，使用如下方式传递环境变量：

# ...

client = MultiServerMCPClient({
    "mysql": {
        "command": "uv",
        "args": ["run", "/home/paul/Documents/mysql-mcp/server.py"],
        "transport": "stdio",
        "env": {
            "MYSQL_HOST": "https://mysql_ip:3306",
            "DB_NAME": "testdb",
            "MYSQL_USER": "root",
            "MYSQL_PASSWORD": "654321"
        },
    }
})

# ...