Spring AI 当中对应 MCP 的操作

启动程序--->读取mcpjson--->通过ProcessBuilder启动命令---> 写入初始化jsonrpc---->写入获取tools列表jsonrpc---->请求大模型（携带tools）---->写入请求外部tool的jsonrpc---->获取数据--->发送给大模型---->响应。在做MCP企业级方案落地时，我们可能不想让没有权限的人访问MCP Server, 或者需要根据不同

casfasfasfa

319人浏览 · 2025-10-01 17:22:02

casfasfasfa · 2025-10-01 17:22:02 发布

MCP

问题：

当有服务商需要将tools提供外部使用（比如高德地图提供了位置服务tools，比如百度提供了联网搜索的tools...）
或者在企业级中，有多个智能应用，想将通用的tools公共化

怎么办？

可以把tools单独抽取出来，由应用程序读取外部的tools。那关键是怎么读呢？怎么解析呢？如果每个提供商各用一种规则你能想象有多麻烦！所以MCP就诞生了，他指定了标准规则，以jsonrpc2.0的方式进行通讯。

那问题又来了，以什么方式通讯呢？ http? rpc? stdio? mcp提供了sse和stdio这2种方式。

使用

Streamable http目前springai1.0版本不支持（因为Streamable http 是 spring ai 1.0 之后说明的）我们先掌握SSE和STDIO

分别说下STDIO和SSE的方式：

STDIO更适合客户端桌面应用和辅助工具
SSE更适合web应用、业务有关的公共tools

MCP STDIO 输出配置实操

MCP Server

现成共用MCP Server

现在有很多MCP 服务给大家提供一个网站：MCP Server（MCP 服务器）

那MCP有了，怎么调用呢？这里介绍2种使用方式：

MCP Client

通过工具

CherryStudio、Cursor 、Claude Desktop、Cline 等等很多，这里不一一演示，不会的话自己找个文章，工具使用都很简单!

以Cline为例：他是Vscode的插件

安装VSCode
安装插件：

配置cline的模型：

配置cline的mcpserver

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-json">{
    <span style="color:#015692">"mcpServers"</span>: {
        <span style="color:#015692">"baidu-map"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@baidumap/mcp-server-baidu-map"</span>
            ],
            <span style="color:#015692">"env"</span>: {
                <span style="color:#015692">"BAIDU_MAP_API_KEY"</span>: <span style="color:#54790d">"LEyBQxG9UzR9C1GZ6zDHsFDVKvBem2do"</span>
            }
        },
        <span style="color:#015692">"filesystem"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@modelcontextprotocol/server-filesystem"</span>,
                <span style="color:#54790d">"C:/Users/tuling/Desktop"</span>
            ]
        },
        <span style="color:#015692">"mcp-server-weather"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"java"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>,
                <span style="color:#54790d">"-jar"</span>,
                <span style="color:#54790d">"D:\\ideaworkspace\\git_pull\\tuling-flight-booking_all\\mcp-stdio-server\\target\\mcp-stdio-server-xs-1.0.jar"</span>
            ]
        }
    }
}
</code></span></span>

开启cline权限

6.测试：

通过 Spring AI 接入第三方的 MCP Server

依赖

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-xml"><span style="color:#656e77"><!--既支持sse\也支持Stdio--></span>
<<span style="color:#b75501">dependency</span>>
    <<span style="color:#b75501">groupId</span>>org.springframework.ai</<span style="color:#b75501">groupId</span>>
    <<span style="color:#b75501">artifactId</span>>spring-ai-starter-mcp-client-webflux</<span style="color:#b75501">artifactId</span>>
</<span style="color:#b75501">dependency</span>>
</code></span></span>

2 配置

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-yaml"><span style="color:#015692">spring:</span>
  <span style="color:#015692">ai:</span>
    <span style="color:#015692">mcp:</span>
      <span style="color:#015692">client:</span>
      <span style="color:#656e77"># 连接超时时间设置</span>
        <span style="color:#015692">request-timeout:</span> <span style="color:#b75501">60000</span> 
        <span style="color:#015692">stdio:</span> <span style="color:#656e77"># 设置 sse 输出方式</span>
        <span style="color:#656e77"># 配置Mcp 方式2： 将 mcp的配置 单独放在一个 Json 文件当中读取，推荐，利用维护</span>
        <span style="color:#656e77"># classpath 是指:项目resources</span>
          <span style="color:#015692">servers-configuration:</span> <span style="color:#54790d">classpath:/mcp-servers-config.json</span>
        <span style="color:#656e77"># 配置MCP 方式2： 直接将 mcp 配置全局配置文件中（mcp 配置太多不利于维护）</span>
          <span style="color:#656e77"># connections:</span>
          <span style="color:#656e77">#   server1:</span>
          <span style="color:#656e77">#     command: /path/to/server</span>
          <span style="color:#656e77">#     args:</span>
          <span style="color:#656e77">#       - --port=8080</span>
          <span style="color:#656e77">#       - --mode=production</span>
          <span style="color:#656e77">#     env:</span>
          <span style="color:#656e77">#       API_KEY: your-api-key</span>
          <span style="color:#656e77">#       DE<span style="color:#015692">BUG:</span> "true"</span>
</code></span></span>

mcp-servers-config.json：

获取Baidu地图key: 控制台 | 百度地图开放平台

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-json">{
    <span style="color:#015692">"mcpServers"</span>: {
        <span style="color:#015692">"baidu-map"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@baidumap/mcp-server-baidu-map"</span>
            ],
            <span style="color:#015692">"env"</span>: {
                <span style="color:#015692">"BAIDU_MAP_API_KEY"</span>: <span style="color:#54790d">"xxxx"</span>
            }
        },
        <span style="color:#015692">"filesystem"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@modelcontextprotocol/server-filesystem"</span>,
                <span style="color:#54790d">"C:/Users/tuling/Desktop"</span>
            ]
        },
        <span style="color:#015692">"mcp-server-weather"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"java"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>,
                <span style="color:#54790d">"-jar"</span>,
                <span style="color:#54790d">"D:\\xxx\\target\\mcp-stdio-server-xs-1.0.jar"</span>
            ]
        }
    }
}
</code></span></span>

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-json">{
    <span style="color:#015692">"mcpServers"</span>: {
      <span style="color:#656e77">// 外部第三方的</span>
        <span style="color:#015692">"baidu-map"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@baidumap/mcp-server-baidu-map"</span>
            ],
            <span style="color:#015692">"env"</span>: {
                <span style="color:#015692">"BAIDU_MAP_API_KEY"</span>: <span style="color:#54790d">"xxxx"</span>
            }
        },
       <span style="color:#656e77">// 外部第三方的</span>
        <span style="color:#015692">"filesystem"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,  <span style="color:#656e77">// 指明使用 cmd 命令执行</span>
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@modelcontextprotocol/server-filesystem"</span>,
                <span style="color:#54790d">"C:/Users/tuling/Desktop"</span>
            ]
        },
       <span style="color:#656e77">// 自定义的 mcp 服务</span>
        <span style="color:#015692">"mcp-server-weather"</span>: {  <span style="color:#656e77">// 对应的项目名 application的 name</span>
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"java"</span>, <span style="color:#656e77">// 指明通过 java 命令执行，java 解析可以直接识别到</span>
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>, <span style="color:#656e77">// 清空控制台,不然会输入很多信息</span>
                <span style="color:#54790d">"-jar"</span>, <span style="color:#656e77">// -jar 启动 Spring Boot</span>
                <span style="color:#54790d">"D:\\xxx\\target\\mcp-stdio-server-xs-1.0.jar"</span> <span style="color:#656e77">// 自定义的mcp服务的jar路径</span>
            ]
        }
    }
}
</code></span></span>

绑定到Chatclient

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#656e77">/**
 * <span style="color:#015692">@description</span>: 智能航空助手:
 */</span>
<span style="color:#015692">@RestController</span>
<span style="color:#015692">@CrossOrigin</span>
<span style="color:#015692">public</span> <span style="color:#015692">class</span> <span style="color:#b75501">OpenAiController</span> {
    
    <span style="color:#015692">private</span> <span style="color:#015692">final</span> ChatClient chatClient;
    
    <span style="color:#015692">public</span> <span style="color:#b75501">OpenAiController</span>(
            DashScopeChatModel dashScopeChatModel,
                            // 配置引入 外部 mcp tools
                            ToolCallbackProvider mcpTools) {
        <span style="color:#b75501">this</span>.chatClient =ChatClient.builder(dashScopeChatModel)
        .defaultToolCallbacks(mcpTools)  <span style="color:#656e77">// 将外部的 mcop tools 对大模型进行绑定,这里是构造器的绑定，不是单个对话的绑定</span>
        .build();
    }
    

 <span style="color:#015692">@CrossOrigin</span>
<span style="color:#015692">@GetMapping(value = "/ai/generateStreamAsString", produces = MediaType.TEXT_EVENT_STREAM_VALUE)</span>
<span style="color:#015692">public</span> Flux<String> <span style="color:#b75501">generateStreamAsString</span>(<span style="color:#015692">@RequestParam(value = "message", defaultValue = "讲个笑话")</span> String message) {

    Flux<String> content = chatClient.prompt()
            .user(message)
            .stream()
            .content();

    <span style="color:#015692">return</span>  content;

    }
</code></span></span>

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-yaml"><span style="color:#656e77"># 调试日志</span>
<span style="color:#015692">logging:</span>
  <span style="color:#015692">level:</span>
    <span style="color:#015692">io:</span>
      <span style="color:#015692">modelcontextprotocol:</span>
        <span style="color:#015692">client:</span> <span style="color:#54790d">DEBUG</span>
        <span style="color:#015692">spec:</span> <span style="color:#54790d">DEBUG</span>
</code></span></span>

使用 Spring AI 接入自定义MCP Server

创建一个spring ai项目

依赖

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-xml"><span style="color:#656e77"><!--mcp-server  --></span>
<<span style="color:#b75501">dependency</span>>
  <<span style="color:#b75501">groupId</span>>org.springframework.ai</<span style="color:#b75501">groupId</span>>
  <<span style="color:#b75501">artifactId</span>>spring-ai-starter-mcp-server</<span style="color:#b75501">artifactId</span>>
</<span style="color:#b75501">dependency</span>>

 <<span style="color:#b75501">dependencyManagement</span>>
        <<span style="color:#b75501">dependencies</span>>
            <span style="color:#656e77"><!--spring ai 包管理依赖 --></span>
            <<span style="color:#b75501">dependency</span>>
                <<span style="color:#b75501">groupId</span>>org.springframework.ai</<span style="color:#b75501">groupId</span>>
                <<span style="color:#b75501">artifactId</span>>spring-ai-bom</<span style="color:#b75501">artifactId</span>>
                <<span style="color:#b75501">version</span>>${spring-ai.version}</<span style="color:#b75501">version</span>>
                <<span style="color:#b75501">type</span>>pom</<span style="color:#b75501">type</span>>
                <<span style="color:#b75501">scope</span>>import</<span style="color:#b75501">scope</span>>
            </<span style="color:#b75501">dependency</span>>
        </<span style="color:#b75501">dependencies</span>>
  </<span style="color:#b75501">dependencyManagement</span>>

<span style="color:#656e77"><!-- 打包 --></span>
<<span style="color:#b75501">build</span>>
        <<span style="color:#b75501">plugins</span>>
            <<span style="color:#b75501">plugin</span>>
                <<span style="color:#b75501">groupId</span>>org.springframework.boot</<span style="color:#b75501">groupId</span>>
                <<span style="color:#b75501">artifactId</span>>spring-boot-maven-plugin</<span style="color:#b75501">artifactId</span>>
                <<span style="color:#b75501">executions</span>>
                    <<span style="color:#b75501">execution</span>>
                        <<span style="color:#b75501">goals</span>>
                            <<span style="color:#b75501">goal</span>>repackage</<span style="color:#b75501">goal</span>>
                        </<span style="color:#b75501">goals</span>>
                    </<span style="color:#b75501">execution</span>>
                </<span style="color:#b75501">executions</span>>
            </<span style="color:#b75501">plugin</span>>
        </<span style="color:#b75501">plugins</span>>
    </<span style="color:#b75501">build</span>>
</code></span></span>

添加工具

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#015692">@Service</span>
<span style="color:#015692">public</span> <span style="color:#015692">class</span> <span style="color:#b75501">UserToolService</span> {

    Map<String,Double> userScore = Map.of(
        <span style="color:#54790d">"xushu"</span>,<span style="color:#b75501">99.0</span>,
        <span style="color:#54790d">"zhangsan"</span>,<span style="color:#b75501">2.0</span>,
        <span style="color:#54790d">"lisi"</span>,<span style="color:#b75501">3.0</span>);
    <span style="color:#015692">@Tool(description = "获取用户分数")</span>
    <span style="color:#015692">public</span> String <span style="color:#b75501">getScore</span>(String username) { <span style="color:#656e77">// 也可以添加上 @ToolParam(description=“” )告诉大模型这个参数的描述是做什么的</span>
        <span style="color:#015692">if</span>(userScore.containsKey(userName)){
            <span style="color:#015692">return</span> userScore.get(userName).toString();
        }  

        <span style="color:#015692">return</span> <span style="color:#54790d">"未检索到当前用户"</span>+userName;
    }
}
</code></span></span>

暴露工具

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#015692">@Bean</span>  <span style="color:#656e77">// 将我们编写的 tools 对外的UserToolService 绑定上去</span>
<span style="color:#015692">public</span> ToolCallbackProvider <span style="color:#b75501">weatherTools</span>(UserToolService userToolService) {
    <span style="color:#015692">return</span> MethodToolCallbackProvider.builder().toolObjects(userToolService).build();
}
</code></span></span>

配置

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-yaml"><span style="color:#015692">spring:</span>
  <span style="color:#015692">main:</span>
    <span style="color:#015692">banner-mode:</span> <span style="color:#54790d">off</span>
  <span style="color:#015692">ai:</span>
    <span style="color:#015692">mcp:</span>
      <span style="color:#015692">server:</span>
        <span style="color:#015692">name:</span> <span style="color:#54790d">my-weather-server</span>
        <span style="color:#015692">version:</span> <span style="color:#b75501">0.0</span><span style="color:#b75501">.1</span>
</code></span></span>

# 注意：您必须禁用横幅和控制台日志记录，以允许 STDIO 传输!!工作 banner-mode: off

打包 mvn package

此时target/生成了jar则成功！

在我们需要的用到我们自定义的 mcp 的项目当中，加上我们自行定义的 MCP 服务。如下，我们是将其统一放到了一个配置的 json 文件当中。去了

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-json">{
    <span style="color:#015692">"mcpServers"</span>: {
      <span style="color:#656e77">// 外部第三方的</span>
        <span style="color:#015692">"baidu-map"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@baidumap/mcp-server-baidu-map"</span>
            ],
            <span style="color:#015692">"env"</span>: {
                <span style="color:#015692">"BAIDU_MAP_API_KEY"</span>: <span style="color:#54790d">"xxxx"</span>
            }
        },
       <span style="color:#656e77">// 外部第三方的</span>
        <span style="color:#015692">"filesystem"</span>: {
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,  <span style="color:#656e77">// 指明使用 cmd 命令执行</span>
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"/c"</span>,
                <span style="color:#54790d">"npx"</span>,
                <span style="color:#54790d">"-y"</span>,
                <span style="color:#54790d">"@modelcontextprotocol/server-filesystem"</span>,
                <span style="color:#54790d">"C:/Users/tuling/Desktop"</span>
            ]
        },
       <span style="color:#656e77">// 自定义的 mcp 服务</span>
        <span style="color:#015692">"mcp-server-weather"</span>: {  <span style="color:#656e77">// 对应的项目名 application的 name</span>
            <span style="color:#015692">"command"</span>: <span style="color:#54790d">"java"</span>, <span style="color:#656e77">// 指明通过 java 命令执行，java 解析可以直接识别到</span>
            <span style="color:#015692">"args"</span>: [
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>, <span style="color:#656e77">// 清空控制台,不然会输入很多信息</span>
                <span style="color:#54790d">"-jar"</span>, <span style="color:#656e77">// -jar 启动 Spring Boot</span>
                <span style="color:#54790d">"D:\\xxx\\target\\mcp-stdio-server-xs-1.0.jar"</span> <span style="color:#656e77">// 自定义的mcp服务的jar路径</span>
            ]
        }
    }
}
</code></span></span>

MCP SSE 输出配置实操（推荐 Web）

MCP Server

这种方式需要将部署为Web服务

依赖

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-xml">      <span style="color:#656e77"><!--mcp服务器核心依赖— 响应式--></span>
      <<span style="color:#b75501">dependency</span>>
        <<span style="color:#b75501">groupId</span>>org.springframework.ai</<span style="color:#b75501">groupId</span>>
        <<span style="color:#b75501">artifactId</span>>spring-ai-starter-mcp-server-webflux</<span style="color:#b75501">artifactId</span>>
      </<span style="color:#b75501">dependency</span>>
      <span style="color:#656e77"><!-- 这个 SSE 是需要 Web 的 --></span>
      <<span style="color:#b75501">dependency</span>>
        <<span style="color:#b75501">groupId</span>>org.springframework</<span style="color:#b75501">groupId</span>>
        <<span style="color:#b75501">artifactId</span>>spring-boot-starter-web</<span style="color:#b75501">artifactId</span>>
      </<span style="color:#b75501">dependency</span>>
</code></span></span>

定义外部工具

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#015692">@Service</span>
<span style="color:#015692">public</span> <span style="color:#015692">class</span> <span style="color:#b75501">UserToolService</span> {

    Map<String,Double> userScore = Map.of(
            <span style="color:#54790d">"xushu"</span>,<span style="color:#b75501">99.0</span>,
            <span style="color:#54790d">"zhangsan"</span>,<span style="color:#b75501">2.0</span>,
            <span style="color:#54790d">"lisi"</span>,<span style="color:#b75501">3.0</span>);
    <span style="color:#015692">@Tool(description = "获取用户分数")</span>
    <span style="color:#015692">public</span> String <span style="color:#b75501">getScore</span>(String username) {
        <span style="color:#015692">if</span>(userScore.containsKey(username)){
            <span style="color:#015692">return</span> userScore.get(username).toString();
        }

        <span style="color:#015692">return</span> <span style="color:#54790d">"未检索到当前用户"</span>;
    }
}
</code></span></span>

暴露工具

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#015692">@Bean</span>
    <span style="color:#015692">public</span> ToolCallbackProvider <span style="color:#b75501">weatherToolCallbackProvider</span>(WeatherService weatherService,
                                                            UserToolService userToolService) {
        <span style="color:#015692">return</span> MethodToolCallbackProvider.builder().toolObjects(userToolService).build();
    }
</code></span></span>

配置（需要用 web 启动）

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-yaml"><span style="color:#015692">server:</span>
  <span style="color:#015692">port:</span> <span style="color:#b75501">8088</span>
</code></span></span>

MCP Client

将上面通过 SSE 方式创建的自定义 MCP Server 配置进来

添加依赖

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-xml"><span style="color:#656e77"><!--既支持sse\也支持Stdio--></span>
<<span style="color:#b75501">dependency</span>>
  <<span style="color:#b75501">groupId</span>>org.springframework.ai</<span style="color:#b75501">groupId</span>>
  <<span style="color:#b75501">artifactId</span>>spring-ai-starter-mcp-client-webflux</<span style="color:#b75501">artifactId</span>>
</<span style="color:#b75501">dependency</span>>
</code></span></span>

配置

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-yaml"><span style="color:#015692">spring:</span>
  <span style="color:#015692">ai:</span>
    <span style="color:#015692">mcp:</span>
      <span style="color:#015692">client:</span>
        <span style="color:#015692">enabled:</span> <span style="color:#b75501">true</span>
        <span style="color:#015692">name:</span> <span style="color:#54790d">my-mcp-client</span>
        <span style="color:#015692">version:</span> <span style="color:#b75501">1.0</span><span style="color:#b75501">.0</span>
        <span style="color:#015692">request-timeout:</span> <span style="color:#54790d">30s</span>
        <span style="color:#015692">type:</span> <span style="color:#54790d">ASYNC</span>  <span style="color:#656e77"># or SYNC</span>
        <span style="color:#015692">sse:</span> <span style="color:#656e77"># 设置 sse 输出方式</span>
          <span style="color:#015692">connections:</span>
            <span style="color:#015692">server1:</span>
              <span style="color:#015692">url:</span> <span style="color:#54790d">http://localhost:8088</span>
</code></span></span>

代码

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#656e77">/**
 * <span style="color:#015692">@author</span> wx:程序员徐庶
 * <span style="color:#015692">@version</span> 1.0
 * <span style="color:#015692">@description</span>: 智能航空助手:需要一对一解答关注wx: 程序员徐庶
 */</span>
<span style="color:#015692">@RestController</span>
<span style="color:#015692">@CrossOrigin</span>
<span style="color:#015692">public</span> <span style="color:#015692">class</span> <span style="color:#b75501">OpenAiController</span> {

    <span style="color:#015692">private</span> <span style="color:#015692">final</span> ChatClient chatClient;

    <span style="color:#015692">public</span> <span style="color:#b75501">OpenAiController</span>(
        DashScopeChatModel dashScopeChatModel,
        // 外部 mcp tools
        ToolCallbackProvider mcpTools) {
        <span style="color:#b75501">this</span>.chatClient =ChatClient.builder(dashScopeChatModel)
        .defaultToolCallbacks(mcpTools)
        .build();
    }


    <span style="color:#015692">@CrossOrigin</span>
    <span style="color:#015692">@GetMapping(value = "/ai/generateStreamAsString", produces = MediaType.TEXT_EVENT_STREAM_VALUE)</span>
    <span style="color:#015692">public</span> Flux<String> <span style="color:#b75501">generateStreamAsString</span>(<span style="color:#015692">@RequestParam(value = "message", defaultValue = "讲个笑话")</span> String message) {

        Flux<String> content = chatClient.prompt()
        .user(message)
        .stream()
        .content();

        <span style="color:#015692">return</span>  content;

    }
</code></span></span>

原理

STDIO 是基于标准输入\输出流的方式，需要在MCP 客户端安装一个包（可以是jar包、python包、npm包等..）. 它是“客户端”的MCP Server。

SSE 是基于Http的方式进行通讯，需要将MCP Server部署为一个web服务. 它是服务端的MCP Server

STDIO原理

很多人不理解stdio到底什么意思，为什么一定要把stdio server的banner关掉，还要清空控制台？

首先SpringAi底层会读取到mcp-servers-config.json的信息
然后执行命令（其实聪明的小伙伴早就发现了，mcp-servers-config.json文件中就是一堆shell命令）
1. 怎么执行？熟悉java的同学应该知道，java里面有一个对象用于执行命令：

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#b75501">ProcessBuilder</span> <span style="color:#54790d">processBuilder</span> <span style="color:#ab5656">=</span> <span style="color:#015692">new</span> <span style="color:#b75501">ProcessBuilder</span>();
        processBuilder.command(<span style="color:#54790d">"java"</span>,<span style="color:#54790d">"-version"</span>);

        <span style="color:#b75501">Process</span> <span style="color:#54790d">process</span> <span style="color:#ab5656">=</span> processBuilder.start();

        process.errorReader().lines().forEach(System.out::println);
</code></span></span>

所以springAi底层相当于读取到信息后，会通过processBuilder去执行命令

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java">String[] commands={<span style="color:#54790d">"java"</span>,
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>,
                <span style="color:#54790d">"-jar"</span>,
                <span style="color:#54790d">"D:\\ideaworkspace\\git_pull\\tuling-flight-booking_all\\mcp-stdio-server\\target\\mcp-stdio-server-xs-1.0.jar"</span>};

        <span style="color:#b75501">ProcessBuilder</span> <span style="color:#54790d">processBuilder</span> <span style="color:#ab5656">=</span> <span style="color:#015692">new</span> <span style="color:#b75501">ProcessBuilder</span>();
        processBuilder.command(commands);
        <span style="color:#656e77">// processBuilder.environment().put("username","xushu");</span>

        <span style="color:#b75501">Process</span> <span style="color:#54790d">process</span> <span style="color:#ab5656">=</span> processBuilder.start();
</code></span></span>

其实你也完全可以自己通过mcd去执行命令

运行jar -jar mcp-stdio-server.jar
输入{"jsonrpc":"2.0","method":"tools/list","id":"3b3f3431-1","params":{}}
输出tools列表

这就是标准输入输出流! 看到这里你应该知道，为什么需要-Dlogging.pattern.console= 完全是为了清空控制台，才能读取信息!

所以利用java也是一样的原理：

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-java"><span style="color:#015692">@Test</span>
    <span style="color:#015692">public</span> <span style="color:#015692">void</span> <span style="color:#b75501">test</span>() <span style="color:#015692">throws</span> IOException, InterruptedException {
        String[] commands={<span style="color:#54790d">"java"</span>,
                <span style="color:#54790d">"-Dspring.ai.mcp.server.stdio=true"</span>,
                <span style="color:#54790d">"-Dlogging.pattern.console="</span>,
                <span style="color:#54790d">"-jar"</span>,
                <span style="color:#54790d">"D:\\ideaworkspace\\git_pull\\tuling-flight-booking_all\\mcp-stdio-server\\target\\mcp-stdio-server-xs-1.0.jar"</span>};

        <span style="color:#b75501">ProcessBuilder</span> <span style="color:#54790d">processBuilder</span> <span style="color:#ab5656">=</span> <span style="color:#015692">new</span> <span style="color:#b75501">ProcessBuilder</span>();
        processBuilder.command(commands);
        processBuilder.environment().put(<span style="color:#54790d">"username"</span>,<span style="color:#54790d">"xushu"</span>);

        <span style="color:#b75501">Process</span> <span style="color:#54790d">process</span> <span style="color:#ab5656">=</span> processBuilder.start();

        <span style="color:#b75501">Thread</span> <span style="color:#54790d">thread</span> <span style="color:#ab5656">=</span> <span style="color:#015692">new</span> <span style="color:#b75501">Thread</span>(() -> {
            <span style="color:#015692">try</span> (<span style="color:#b75501">BufferedReader</span> <span style="color:#54790d">processReader</span> <span style="color:#ab5656">=</span> <span style="color:#015692">new</span> <span style="color:#b75501">BufferedReader</span>(<span style="color:#015692">new</span> <span style="color:#b75501">InputStreamReader</span>(process.getInputStream()))) {
                String line;
                <span style="color:#015692">while</span> ((line=processReader.readLine())!=<span style="color:#b75501">null</span>) {
                        System.out.println(line);
                }
            } <span style="color:#015692">catch</span> (IOException e) {
                e.printStackTrace();
            }
        });
        thread.start();


        Thread.sleep(<span style="color:#b75501">1000</span>);

        <span style="color:#015692">new</span> <span style="color:#b75501">Thread</span>(() -> {

            <span style="color:#015692">try</span> {
                <span style="color:#656e77">//String jsonMessage="{\"jsonrpc\":\"2.0\",\"method\":\"initialize\",\"id\":\"3670122a-0\",\"params\":{\"protocolVersion\":\"2024-11-05\",\"capabilities\":{},\"clientInfo\":{\"name\":\"spring-ai-mcp-client\",\"version\":\"1.0.0\"}}}";</span>
                <span style="color:#b75501">String</span> <span style="color:#54790d">jsonMessage</span> <span style="color:#ab5656">=</span> <span style="color:#54790d">"{\"jsonrpc\":\"2.0\",\"method\":\"tools/list\",\"id\":\"3b3f3431-1\",\"params\":{}}"</span>;

                jsonMessage = jsonMessage.replace(<span style="color:#54790d">"\r\n"</span>, <span style="color:#54790d">"\\n"</span>).replace(<span style="color:#54790d">"\n"</span>, <span style="color:#54790d">"\\n"</span>).replace(<span style="color:#54790d">"\r"</span>, <span style="color:#54790d">"\\n"</span>);

                <span style="color:#b75501">var</span> <span style="color:#54790d">os</span> <span style="color:#ab5656">=</span> process.getOutputStream();
                <span style="color:#015692">synchronized</span> (os) {
                    os.write(jsonMessage.getBytes(StandardCharsets.UTF_8));
                    os.write(<span style="color:#54790d">"\n"</span>.getBytes(StandardCharsets.UTF_8));
                    os.flush();
                }
                System.out.println(<span style="color:#54790d">"写入完成！"</span>);
            }<span style="color:#015692">catch</span> (IOException e){
                e.printStackTrace();
            }
        }).start();


        thread.join();
        <span style="color:#656e77">/*JSONRPCRequest[jsonrpc=2.0, method=initialize, id=5d83d0d1-0, params=InitializeRequest[protocolVersion=2024-11-05, capabilities=ClientCapabilities[experimental=null, roots=null, sampling=null],
        clientInfo=Implementation[name=spring-ai-mcp-client, version=1.0.0]]]*/</span>
    }
</code></span></span>

通过ProcessBuilder执行命令
通过子线程轮询 process.getInputStream 获取输出流
通过process.getOutputStream(); 进行写入流

所以整个过程是这样的：再回顾上面的图

启动程序--->读取mcpjson--->通过ProcessBuilder启动命令---> 写入初始化jsonrpc---->写入获取tools列表jsonrpc---->请求大模型（携带tools）---->写入请求外部tool的jsonrpc---->获取数据--->发送给大模型---->响应。

STDIO源码

MCP鉴权

在做MCP企业级方案落地时，我们可能不想让没有权限的人访问MCP Server, 或者需要根据不同的用户返回不同的数据，这里就涉及到MCP Server授权操作。

那MCP Server有2种传输方式，实现起来不一样：

STDIO

这种方式在本地运行,它将MCP Server作为子进程启动。我们称为标准输入输出，其实就是利用运行命令的方式写入和读取控制台的信息，以达到传输。

通常我们会配置一段json，比如这里的百度地图MCP Server ：

其中command和args代表运行的命令和参数。
其实env中的节点BAIDU_MAP_API_KEY就是做授权的。

如果你传入的BAIDU_MAP_API_KEY不对，就没有使用权限。

<span style="color:#000000"><span style="background-color:#fefef2"><code class="language-json"><span style="color:#015692">"baidu-map"</span>: {
  <span style="color:#015692">"command"</span>: <span style="color:#54790d">"cmd"</span>,
  <span style="color:#015692">"args"</span>: [
    <span style="color:#54790d">"/c"</span>,
    <span style="color:#54790d">"npx"</span>,
    <span style="color:#54790d">"-y"</span>,
    <span style="color:#54790d">"@baidumap/mcp-server-baidu-map"</span>
  ],
  <span style="color:#015692">"env"</span>: {
    <span style="color:#015692">"BAIDU_MAP_API_KEY"</span>: <span style="color:#54790d">"LEyBQxG9UzR9C1GZ6zDHsFDVKvBem2do"</span>
  }
},
</code></span></span>

所以STDIO做授权的方式很明确，就是通过env【环境变量】，实现步骤如下：

服务端发放一个用户的凭证（可以是秘钥、token）这步不细讲，需要有一个授权中心发放凭证。
通过mcp client通过env传入凭证
mcp server通过环境变量鉴权

所以在MCP Server端就可以通过获取环境变量的方式获取env里面的变量：

也可以通过AOP的方式统一处理

北京朝阳AI社区

更多推荐

2025年人工智能状况：市场领导者、企业应用与下一前沿

2025年生成式AI市场呈现两极分化格局：少数巨头平台（如ChatGPT、Gemini、Claude）主导通用市场，而专业化工具（如Midjourney、GitHub Copilot）在垂直领域蓬勃发展。市场评估标准转向ROI、数据安全和合规性。多模态AI和代理式AI成为核心趋势，前者实现跨媒体交互，后者完成自主任务执行。开源与闭源模型之争加剧，同时全球AI法规推动合规性成为战略核心。中国市场由本