实现 MCP Server 开发指南

本文系统性讲解如何基于 tio-boot 从零实现一个符合 MCP（Model Context Protocol） 的 Server，并通过 SSE（Server-Sent Events） 与 MCP Client（如 openai-mcp）完成真实协议交互。文末保留了完整代码与完整抓包协议内容，便于对照实现细节与运行行为。

---

• 一、MCP 协议概览
  • 1. MCP 是什么
  • 2. MCP 的关键约定
• 二、交互与通信模型：HTTP + JSON-RPC + SSE
• 三、基于 tio-boot 的整体架构设计
• 四、核心实现：McpHandler（协议解析与 SSE 输出）
  • 1. 关键职责
  • 2. SSE 的 tio-boot 实现方式
• 五、业务实现：McpCoderServer（Tool 注册与执行）
  • 1. Tool schema 的价值
  • 2. 示例工具
• 六、路由挂载与启动方式
• 七、完整请求 / 响应流程复盘（结合真实抓包）
• 八、常见注意事项与排错建议
• 1）McpHandler
• 2）McpCoderServer
• 3）路由挂载
• mcp协议解析
  • 1.initialize and tool list
  • 2.tool call

---

一、MCP 协议概览

1. MCP 是什么

MCP（Model Context Protocol） 是一套用于 AI Client ↔ Tool Server 的协议标准，目标是让不同模型/客户端以统一方式发现并调用工具服务。协议核心点：

• 统一会话与能力协商：initialize
• 统一工具发现：tools/list
• 统一工具调用：tools/call
• 统一生命周期结束：DELETE
• 支持流式返回：常用 SSE 将 JSON-RPC 响应以事件流方式写回

2. MCP 的关键约定

项	说明
通信协议	HTTP + JSON-RPC 2.0
返回形式	SSE（text/event-stream，常见 Transfer-Encoding: chunked）
会话标识	mcp-session-id（由 Server 返回，Client 后续携带）
版本协商	mcp-protocol-version / Mcp-Protocol-Version
生命周期	initialize → notifications/initialized → tools/list → tools/call → DELETE

---

二、交互与通信模型：HTTP + JSON-RPC + SSE

在 MCP 中，请求通常为：

• HTTP POST 到 /mcp

• Body 为 JSON-RPC 2.0 格式（jsonrpc/method/id/params）

• Server 对某些方法（如 initialize、tools/list、tools/call）使用 SSE 返回：

  • 先写出响应头：Content-Type: text/event-stream
  • 然后以 event: message + data: {json} 形式发送 JSON-RPC 响应
  • 连接可保持不断开（用于后续事件/流式）

而对于 notifications/initialized 这种通知型方法，常见做法是：

• 不走 SSE
• 直接返回 202 Accepted（无 body），表示已接收处理

会话关闭通常通过：

• HTTP DELETE /mcp
• 携带 Mcp-Session-Id、Mcp-Protocol-Version 等 header
• Server 执行 sessionClosed(...) 清理资源

---

三、基于 tio-boot 的整体架构设计

┌────────────┐
│ MCP Client │ (openai-mcp)
└─────┬──────┘
      │ HTTP POST / DELETE
      ▼
┌──────────────┐
│ tio-boot     │
│ HttpServer   │
└─────┬────────┘
      ▼
┌────────────────────────┐
│ McpHandler              │  ← 协议解析 / SSE
└─────┬──────────────────┘
      ▼
┌────────────────────────┐
│ McpServer / McpCoder   │  ← 业务逻辑
│ ToolRegistry           │
└────────────────────────┘

核心分工：

• McpHandler：协议入口层

  • 解析 HTTP 请求与 JSON-RPC
  • 管理 mcp-session-id 与 protocolVersion
  • 对需要 SSE 的方法写出 SSE header，并以 chunk 形式推送事件数据
  • 调用真正的业务 Server（IMcpServer）

• McpCoderServer：业务实现层

  • 声明 Server 信息（name/version）
  • 注册 Tools（schema + handler）
  • 实现工具执行逻辑（run_shell / run_python / ai_search 等）

---

四、核心实现：McpHandler（协议解析与 SSE 输出）

1. 关键职责

McpHandler 实现了 tio-boot 的 HttpRequestHandler，负责：

• 从 request header 中读取：

  • mcp-session-id
  • mcp-protocol-version

• 若未提供 sessionId，则生成一个并写回 response header

• 根据 HTTP method：

  • DELETE：执行 sessionClosed
  • POST：解析 JSON-RPC 并按 method 分派

• 对 initialize / tools/list / tools/call：

  • 写出 SSE header（text/event-stream、chunked、keep-alive）
  • 通过 SseEmitter.pushSSEChunk(...) 推送 event: message + data: ...

• 对 notifications/initialized：

  • 调用业务方法并返回 202

2. SSE 的 tio-boot 实现方式

在 tio-boot 里，关键点是：

• 先构造 HttpResponse
• 调用 httpResponse.addServerSentEventsHeader() 设置 SSE 必要头
• 再设置 chunk 传输、keep-alive
• 立刻把 header 发出去（Tio.bSend(channelContext, httpResponse)）
• 同时 httpResponse.setSend(false)，避免 tio 再次自动发送并关闭连接
• 后续用 SseEmitter.pushSSEChunk(...) 不断写入事件数据

这样可以实现 MCP 所需的“建立 SSE 通道并发送 JSON-RPC 响应”。

---

五、业务实现：McpCoderServer（Tool 注册与执行）

McpCoderServer 继承 McpServer，主要做三件事：

1. 声明 MCP Server 的 name/version
2. 注册工具（registry.register(...)）
3. 提供工具 handler（Map args → List&ltMcpContent&gt）

1. Tool schema 的价值

每个 Tool 注册时都定义了：

• name/title/description
• inputSchema（参数类型、required、默认值等）
• handler（执行逻辑）

MCP Client 在 tools/list 时会拿到工具清单 + schema，从而自动生成调用结构。

2. 示例工具

• run_shell：执行 Linux shell 命令，返回 stdout/stderr/exit_code
• run_python：执行 python 代码，返回 stdout/stderr/exit_code
• ai_search：示例 AI 搜索服务（Google + 模型整合），用于演示扩展能力

---

六、路由挂载与启动方式

tio-boot 中把 handler 挂到 /mcp：

McpCoderServer mcpCoderServer = new McpCoderServer();
McpHandler mcpCoderHandler = new McpHandler(mcpCoderServer);
r.add("/mcp", mcpCoderHandler);

当 openai-mcp 访问 POST /mcp 或 DELETE /mcp 时，就会进入 McpHandler.handle(...)。

---

七、完整请求 / 响应流程复盘（结合真实抓包）

下面按抓包顺序解释一次典型会话：

1. Client 发起 initialize

   • POST /mcp

   • Body：{"jsonrpc":"2.0","method":"initialize","id":1,...}

   • Server：

     • 生成或确认 mcp-session-id

     • 写出 SSE header（200 + text/event-stream）

     • SSE 推送：

       • event: message
       • data: {"jsonrpc":"2.0","id":1,"result":{...}}

2. Client 发送 notifications/initialized

   • POST /mcp

   • Body：{"method":"notifications/initialized","jsonrpc":"2.0"}

   • Server：

     • 调用 notificationsInitialized
     • 返回 202 Accepted，content-length 0

3. Client 请求 tools/list

   • POST /mcp

   • Body：{"method":"tools/list","jsonrpc":"2.0","id":1}

   • Server：

     • 写出 SSE header
     • SSE 推送 tools 清单（含 schema）

4. Client 调用 tools/call（抓包里未展示，但逻辑已实现）

   • POST /mcp

   • Body：{"method":"tools/call","id":X,"params":{name,args}}

   • Server：

     • 执行对应 handler
     • SSE 推送 result（contents）

5. Client 结束会话 DELETE

   • DELETE /mcp

   • Server：

     • 调用 sessionClosed
     • 返回 200，content-length 0

---

八、常见注意事项与排错建议

1. SSE header 必须先发送 如果没在 initialize/tools/list/tools/call 里先 Tio.bSend(...)，后续 pushSSEChunk 可能无法写入或连接已关闭。

2. httpResponse.setSend(false) 很关键 不然 tio 可能把 response 当作普通 HTTP 响应处理，发送后关闭连接，SSE 失效。

3. Header 大小写与取值 抓包里 Client 用的是 Mcp-Session-Id、Mcp-Protocol-Version，代码读取的是 mcp-session-id、mcp-protocol-version。多数 HTTP 框架会做大小写不敏感处理，但如果遇到取不到 header，需要确认底层是否规范化了 header 名称。

4. notifications/initialized 返回 202 是合理的 它是通知型 method，不要求 SSE 流。

5. 超时与 keep-aliveKeep-Alive: timeout=60 只是声明，实际还受服务端/代理（如 Cloudflare）影响。生产环境要结合反代/网关配置。

---

附录 A：完整代码（原样保留）

1）McpHandler

package nexus.io.kit.handler;

import java.util.Map;

import com.alibaba.fastjson2.util.TypeUtils;
import nexus.io.mcp.context.McpRequestContext;
import nexus.io.mcp.exception.McpRpcException;
import nexus.io.mcp.model.JsonRpcRequest;
import nexus.io.mcp.model.JsonRpcResponse;
import nexus.io.mcp.model.McpInitializeParams;
import nexus.io.mcp.model.McpInitializeResult;
import nexus.io.mcp.model.McpMethod;
import nexus.io.mcp.model.McpToolsCallParams;
import nexus.io.mcp.model.McpToolsCallResult;
import nexus.io.mcp.model.McpToolsListResult;
import nexus.io.mcp.model.RpcError;
import nexus.io.mcp.server.IMcpServer;
import nexus.io.tio.boot.http.TioRequestContext;
import nexus.io.tio.core.ChannelContext;
import nexus.io.tio.core.Tio;
import nexus.io.tio.http.common.HeaderName;
import nexus.io.tio.http.common.HeaderValue;
import nexus.io.tio.http.common.HttpMethod;
import nexus.io.tio.http.common.HttpRequest;
import nexus.io.tio.http.common.HttpResponse;
import nexus.io.tio.http.common.utils.HttpIpUtils;
import nexus.io.tio.http.server.handler.HttpRequestHandler;
import nexus.io.tio.http.server.util.SseEmitter;
import nexus.io.tio.utils.UUIDUtils;
import nexus.io.tio.utils.hutool.StrUtil;
import nexus.io.tio.utils.json.JsonUtils;

import lombok.extern.slf4j.Slf4j;

@Slf4j
public class McpHandler implements HttpRequestHandler {

  private IMcpServer mcpServer;

  public McpHandler(IMcpServer mcpServer) {
    this.mcpServer = mcpServer;
  }

  @Override
  public HttpResponse handle(HttpRequest httpRequest) {
    HttpResponse httpResponse = TioRequestContext.getResponse();
    ChannelContext channelContext = httpRequest.getChannelContext();

    String mcpSessionId = httpRequest.getHeader("mcp-session-id");
    String protocolVersion = httpRequest.getHeader("mcp-protocol-version");

    if (mcpSessionId == null) {
      mcpSessionId = UUIDUtils.random();
    }
    httpResponse.addHeader("mcp-session-id", mcpSessionId);

    String realIp = HttpIpUtils.getRealIp(httpRequest);
    Map<String, String> headers = httpRequest.getHeaders();

    if (httpRequest.getMethod().equals(HttpMethod.DELETE)) {
      McpRequestContext mcpRequestContext = new McpRequestContext(mcpSessionId, protocolVersion, realIp, headers);
      mcpServer.sessionClosed(mcpRequestContext);
      return httpResponse;
    }

    String bodyString = httpRequest.getBodyString();
    if (StrUtil.isBlank(bodyString)) {
      return httpResponse;
    }

    JsonRpcRequest rpcRequest = JsonUtils.parse(bodyString, JsonRpcRequest.class);
    if (rpcRequest == null) {
      return httpResponse;
    }
    String method = rpcRequest.getMethod();
    Object id = rpcRequest.getId();

    log.info("method:{}", method);
    if (McpMethod.INITIALIZE.equals(method)) {
      sendHttpResponseHeader(httpResponse, channelContext);
      Map<String, Object> params = rpcRequest.getParams();
      McpInitializeParams mcpInitializeParams = TypeUtils.cast(params, McpInitializeParams.class);
      if (protocolVersion == null) {
        protocolVersion = mcpInitializeParams.getProtocolVersion();
      }
      McpRequestContext mcpRequestContext = new McpRequestContext(mcpSessionId, protocolVersion, realIp, headers);

      McpInitializeResult initialize;
      try {
        initialize = mcpServer.initialize(mcpInitializeParams, mcpRequestContext);
        JsonRpcResponse<McpInitializeResult> resp = new JsonRpcResponse<>();
        resp.setId(id).setResult(initialize);
        sendSSEData(channelContext, resp);
      } catch (McpRpcException e) {
        log.error(e.getMessage(), e);
        JsonRpcResponse<?> resp = new JsonRpcResponse<>();
        RpcError rpcError = new RpcError();
        rpcError.setMessage(e.getMessage());
        resp.setId(id).setError(rpcError);
        sendSSEData(channelContext, resp);
      }

    } else if (McpMethod.NOTIFICATIONS_INITIALIZED.equals(method)) {
      McpRequestContext mcpRequestContext = new McpRequestContext(mcpSessionId, protocolVersion, realIp, headers);
      try {
        mcpServer.notificationsInitialized(mcpRequestContext);
        httpResponse.setStatus(202);
      } catch (McpRpcException e) {
        log.error(e.getMessage(), e);
        httpResponse.setStatus(500);
      }

    } else if (McpMethod.TOOLS_LIST.equals(method)) {
      sendHttpResponseHeader(httpResponse, channelContext);

      McpRequestContext mcpRequestContext = new McpRequestContext(mcpSessionId, protocolVersion, realIp, headers);
      McpToolsListResult result;
      try {
        result = mcpServer.listTools(mcpRequestContext);
        JsonRpcResponse<McpToolsListResult> resp = new JsonRpcResponse<>();
        resp.setId(id).setResult(result);
        sendSSEData(channelContext, resp);
      } catch (McpRpcException e) {
        log.error(e.getMessage(), e);
        JsonRpcResponse<?> resp = new JsonRpcResponse<>();
        RpcError rpcError = new RpcError();
        rpcError.setMessage(e.getMessage());
        resp.setId(id).setError(rpcError);
        sendSSEData(channelContext, resp);
      }

    } else if (McpMethod.TOOLS_CALL.equals(method)) {
      sendHttpResponseHeader(httpResponse, channelContext);

      McpRequestContext mcpRequestContext = new McpRequestContext(mcpSessionId, protocolVersion, realIp, headers);

      Map<String, Object> params = rpcRequest.getParams();
      McpToolsCallParams toolCallparams = TypeUtils.cast(params, McpToolsCallParams.class);
      McpToolsCallResult result;
      try {
        result = mcpServer.callTool(toolCallparams, mcpRequestContext);
        JsonRpcResponse<McpToolsCallResult> resp = new JsonRpcResponse<>();
        resp.setId(id).setResult(result);
        sendSSEData(channelContext, resp);
      } catch (McpRpcException e) {
        log.error(e.getMessage(), e);
        JsonRpcResponse<?> resp = new JsonRpcResponse<>();
        RpcError rpcError = new RpcError();
        rpcError.setMessage(e.getMessage());
        resp.setId(id).setError(rpcError);
        sendSSEData(channelContext, resp);
      }
    }
    return httpResponse;
  }

  private void sendSSEData(ChannelContext channelContext, JsonRpcResponse<?> resp) {
    String data = JsonUtils.toSkipNullJson(resp);
    // 发送数据
    SseEmitter.pushSSEChunk(channelContext, "message", data);
    // 手动移除连接
    // SseEmitter.closeChunkConnection(channelContext);
  }

  private void sendHttpResponseHeader(HttpResponse httpResponse, ChannelContext channelContext) {
    // 设置sse请求头
    httpResponse.addServerSentEventsHeader();
    httpResponse.addHeader(HeaderName.Transfer_Encoding, HeaderValue.from("chunked"));
    httpResponse.addHeader(HeaderName.Keep_Alive, HeaderValue.from("timeout=60"));
    // 手动发送消息到客户端,因为已经设置了sse的请求头,所以客户端的连接不会关闭
    Tio.bSend(channelContext, httpResponse);
    httpResponse.setSend(false);
  }
}

2）McpCoderServer

package nexus.io.kit.mcp;

import java.util.ArrayList;
import java.util.List;
import java.util.Map;

import nexus.io.kit.service.GoogleGeminiSearchService;
import nexus.io.kit.utils.CmdInterpreterUtils;
import nexus.io.kit.utils.PythonInterpreterUtils;
import nexus.io.mcp.model.McpContent;
import nexus.io.mcp.model.McpServerInfo;
import nexus.io.mcp.server.McpToolRegistry;
import nexus.io.mcp.server.RegisteredTool;
import nexus.io.model.result.ResultVo;
import nexus.io.tio.utils.commandline.ProcessResult;
import nexus.io.tio.utils.json.JsonUtils;

import lombok.extern.slf4j.Slf4j;

/**
 * 的具体服务器：只需提供名称/版本。 如需添加或覆盖工具，可覆写 registerAdditionalTools() 或单独覆盖某个 handler。
 */
@Slf4j
public class McpCoderServer extends McpServer {

  private GoogleGeminiSearchService aiSearchService = new GoogleGeminiSearchService();

  public McpCoderServer() {
    super();
  }

  @Override
  protected McpServerInfo getMcpServerInfo() {
    return new McpServerInfo("Coder Server", "1.15.0");
  }

  @Override
  protected void registerTools() {
    // run_shell
    RegisteredTool runShell = McpToolRegistry.builder("run_shell").title("run shell")
        .description("执行 Linux Shell 命令并返回 stdout/stderr/exit_code,支持联网").addStringProperty("command", "Command", true)
        .addIntegerProperty("timeout", "Timeout", 30, false).addNullableProperty("cwd", "Cwd", "string", null, false)
        .addNullableObjectProperty("env", "Env", true, null, false).handler(this::handleRunShell).build();
    registry.register(runShell);

    // run_python
    RegisteredTool runPython = McpToolRegistry.builder("run_python").title("run python")
        .description("在子进程中执行一段 donghua代码，返回 stdout/stderr/exit_code,支持联网").addStringProperty("code", "Code", true)
        .addIntegerProperty("timeout", "Timeout", 30, false)
        .addNullableProperty("stdin", "Stdin", "string", null, false)
        .addBooleanProperty("use_isolated", "Use Isolated", true, false).handler(this::handleRunPython).build();
    registry.register(runPython);

    // ai_search（留一个简单的缺省实现，子类可覆盖 handler 或重载注册）
    RegisteredTool aiSearch = McpToolRegistry.builder("ai_search").title("AI Search")
        .description("智能搜索服务。会将问题拆分成多个关键字，通过 Google 搜索获取数据，然后结合模型进行回答,延迟20s.")
        //
        .addStringProperty("question", "Question", true).handler(this::handleAiSearch).build();
    registry.register(aiSearch);

  }

  /** 公共：run_shell 处理逻辑 */
  protected List<McpContent> handleRunShell(Map<String, Object> args) {
    List<McpContent> contents = new ArrayList<>();
    Object command = args.get("command");
    if (command instanceof String) {
      ProcessResult result = CmdInterpreterUtils.executeCmd((String) command);
      String text = JsonUtils.toSkipNullJson(result);
      contents.add(McpContent.buildText(text));
    } else {
      String json = JsonUtils.toSkipNullJson(ResultVo.fail("command must be a string"));
      contents.add(McpContent.buildText(json));
    }
    return contents;
  }

  /** 公共：run_python 处理逻辑 */
  protected List<McpContent> handleRunPython(Map<String, Object> args) {
    List<McpContent> contents = new ArrayList<>();
    Object code = args.get("code");
    if (code instanceof String) {
      ProcessResult result = PythonInterpreterUtils.executeCode((String) code);
      String text = JsonUtils.toSkipNullJson(result);
      contents.add(McpContent.buildText(text));
    } else {
      String json = JsonUtils.toSkipNullJson(ResultVo.fail("code must be a string"));
      contents.add(McpContent.buildText(json));
    }
    return contents;
  }

  /** 公共：ai_search 的缺省实现（示例占位，可被子类替换） */
  protected List<McpContent> handleAiSearch(Map<String, Object> args) {
    List<McpContent> contents = new ArrayList<>();
    Object question = args.get("question");
    if (question instanceof String) {
      log.info("keyword: {}", question);
      try {
        String text = JsonUtils.toSkipNullJson(aiSearchService.search((String) question));
        contents.add(McpContent.buildText(text));
      } catch (Exception e) {
        log.error(e.getMessage(), e);
        String json = JsonUtils.toSkipNullJson(ResultVo.fail(e.getMessage()));
        contents.add(McpContent.buildText(json));
      }

    } else {
      String json = JsonUtils.toSkipNullJson(ResultVo.fail("question must be a string"));
      contents.add(McpContent.buildText(json));
    }
    return contents;
  }
}

3）路由挂载

      McpCoderServer mcpCoderServer = new McpCoderServer();
      McpHandler mcpCoderHandler = new McpHandler(mcpCoderServer);
      r.add("/mcp", mcpCoderHandler);

---

附录 B：完整协议交互内容（原样保留）

mcp协议解析

1.initialize and tool list

1.1.initialize

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 159
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 135.237.133.53
Cf-Ipcountry: US
Cf-Ray: 98726897cbd96ffe-IAD
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Traceparent: 00-68db9731000000000467bf121d36614a-71b754436da77584-00
Tracestate: dd=s:-1;p:2bdd09f56afbf331;t.dm:-3
X-Datadog-Parent-Id: 8977988190380532902
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db973100000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 317432383273328970
X-Forwarded-For: 135.237.133.53, 172.70.174.64
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"jsonrpc":"2.0","method":"initialize","id":1,"params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"openai-mcp","version":"1.0.0"}}}

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:39:15 GMT
server: t-io
cache-control: no-cache, no-transform
connection: keep-alive
content-type: text/event-stream
mcp-session-id: 7e7ab8b566ad4c629abebf920d0cf45b
x-accel-buffering: no
Transfer-Encoding: chunked

event: message
data: {"jsonrpc":"2.0","id":1,"result":{"protocolVersion":"2025-03-26","capabilities":{"experimental":{},"prompts":{"listChanged":false},"resources":{"subscribe":false,"listChanged":false},"tools":{"listChanged":false}},"serverInfo":{"name":"ExecServer","version":"1.15.0"}}}

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 159
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 135.237.133.50
Cf-Ipcountry: US
Cf-Ray: 9872689bae0dd705-IAD
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Traceparent: 00-68db9731000000000467bf121d36614a-328ea6fbd6588e2e-00
Tracestate: dd=s:-1;p:2bdd09f56afbf331;t.dm:-3
X-Datadog-Parent-Id: 6233387827860711166
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db973100000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 317432383273328970
X-Forwarded-For: 135.237.133.50, 172.71.194.46
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"openai-mcp","version":"1.0.0"}},"jsonrpc":"2.0","id":0}

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:39:16 GMT
server: t-io
cache-control: no-cache, no-transform
connection: keep-alive
content-type: text/event-stream
mcp-session-id: 478f1dc036fd49d8ab69a693846acfb6
x-accel-buffering: no
Transfer-Encoding: chunked

event: message
data: {"jsonrpc":"2.0","id":0,"result":{"protocolVersion":"2025-06-18","capabilities":{"experimental":{},"prompts":{"listChanged":false},"resources":{"subscribe":false,"listChanged":false},"tools":{"listChanged":false}},"serverInfo":{"name":"ExecServer","version":"1.15.0"}}}

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 54
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 135.237.133.56
Cf-Ipcountry: US
Cf-Ray: 9872689f3aa0c56b-IAD
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: 478f1dc036fd49d8ab69a693846acfb6
Traceparent: 00-68db9731000000000467bf121d36614a-452f89cdbeb4c92e-00
Tracestate: dd=s:-1;p:2bdd09f56afbf331;t.dm:-3
X-Datadog-Parent-Id: 17731699559690821652
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db973100000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 317432383273328970
X-Forwarded-For: 135.237.133.56, 172.68.15.217
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"method":"notifications/initialized","jsonrpc":"2.0"}

HTTP/1.1 202 Accepted
date: Tue, 30 Sep 2025 08:39:16 GMT
server: t-io
content-type: application/json
mcp-session-id: 478f1dc036fd49d8ab69a693846acfb6
content-length: 0

1.2.tools/list

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 46
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 135.237.133.57
Cf-Ipcountry: US
Cf-Ray: 987268a1cf66d62d-IAD
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: 478f1dc036fd49d8ab69a693846acfb6
Traceparent: 00-68db9731000000000467bf121d36614a-9699cb18b322256d-00
Tracestate: dd=s:-1;p:2bdd09f56afbf331;t.dm:-3
X-Datadog-Parent-Id: 13673633483892079410
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db973100000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 317432383273328970
X-Forwarded-For: 135.237.133.57, 172.70.35.191
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"method":"tools/list","jsonrpc":"2.0","id":1}

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:39:17 GMT
server: t-io
cache-control: no-cache, no-transform
connection: keep-alive
content-type: text/event-stream
mcp-session-id: 478f1dc036fd49d8ab69a693846acfb6
x-accel-buffering: no
Transfer-Encoding: chunked

event: message
data: {"jsonrpc":"2.0","id":1,"result":{"tools":[{"name":"run_shell","title":"run_shell","description":"...... Linux Shell ............... stdout/stderr/exit_code","inputSchema":{"properties":{"command":{"title":"Command","type":"string"},"timeout":{"default":30,"title":"Timeout","type":"integer"},"cwd":{"anyOf":[{"type":"string"},{"type":"null"}],"default":null,"title":"Cwd"},"env":{"anyOf":[{"additionalProperties":true,"type":"object"},{"type":"null"}],"default":null,"title":"Env"}},"required":["command"],"title":"run_shellArguments","type":"object"}},{"name":"run_python","title":"run_python","description":"........................... Python ............... stdout/stderr/exit_code","inputSchema":{"properties":{"code":{"title":"Code","type":"string"},"timeout":{"default":30,"title":"Timeout","type":"integer"},"stdin":{"anyOf":[{"type":"string"},{"type":"null"}],"default":null,"title":"Stdin"},"use_isolated":{"default":true,"title":"Use Isolated","type":"boolean"}},"required":["code"],"title":"run_pythonArguments","type":"object"}}]}}

1.3.Delete

DELETE /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 135.237.133.56
Cf-Ipcountry: US
Cf-Ray: 987268a3cfc3c943-IAD
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: 478f1dc036fd49d8ab69a693846acfb6
Traceparent: 00-68db9731000000000467bf121d36614a-8554ddea46198512-00
Tracestate: dd=s:-1;p:2bdd09f56afbf331;t.dm:-3
X-Datadog-Parent-Id: 9121804070330067398
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db973100000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 317432383273328970
X-Forwarded-For: 135.237.133.56, 172.68.245.99
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:39:17 GMT
server: t-io
content-type: application/json
mcp-session-id: 478f1dc036fd49d8ab69a693846acfb6
content-length: 0

2.tool call

2.1.notifications/initialized

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 54
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 57.154.187.40
Cf-Ipcountry: US
Cf-Ray: 987280c19b39173b-SJC
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: c362f6dd171c4c0b9ee5d230ba011060
Traceparent: 00-68db9b0a00000000e8fb2ae346302db7-6ad9b9d3762b8813-00
Tracestate: dd=s:-1;p:11a91123b3c77b78;t.dm:-3
X-Datadog-Parent-Id: 8650946564966287682
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db9b0a00000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 16788059191577161143
X-Forwarded-For: 57.154.187.40, 172.69.135.128
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"method":"notifications/initialized","jsonrpc":"2.0"}

HTTP/1.1 202 Accepted
date: Tue, 30 Sep 2025 08:55:45 GMT
server: t-io
content-type: application/json
mcp-session-id: c362f6dd171c4c0b9ee5d230ba011060
content-length: 0

2.2.tools/call

POST /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Content-Length: 450
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 57.154.187.36
Cf-Ipcountry: US
Cf-Ray: 987280c27b2467f2-SJC
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: c362f6dd171c4c0b9ee5d230ba011060
Traceparent: 00-68db9b0a00000000e8fb2ae346302db7-b23a3a53d8b444cc-00
Tracestate: dd=s:-1;p:11a91123b3c77b78;t.dm:-3
X-Datadog-Parent-Id: 13928377909303532074
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db9b0a00000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 16788059191577161143
X-Forwarded-For: 57.154.187.36, 172.69.135.128
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

{"method":"tools/call","params":{"_meta":{"openai/userAgent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/140.0.0.0 Safari/537.36","openai/locale":"en-US","openai/userLocation":{"city":"Honolulu","region":"Hawaii","country":"US","timezone":"Pacific/Honolulu","latitude":"21.28870","longitude":"-157.80060"}},"name":"run_python","arguments":{"code":"import sys; print(sys.version)"}},"jsonrpc":"2.0","id":1}

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:55:45 GMT
server: t-io
cache-control: no-cache, no-transform
connection: keep-alive
content-type: text/event-stream
mcp-session-id: c362f6dd171c4c0b9ee5d230ba011060
x-accel-buffering: no
Transfer-Encoding: chunked

event: message
data: {"jsonrpc":"2.0","id":1,"result":{"content":[{"type":"text","text":"{\n  \"exit_code\": 0,\n  \"stdout\": \"3.11.13 | packaged by Anaconda, Inc. | (main, Jun  5 2025, 13:03:15) [MSC v.1929 64 bit (AMD64)]\\r\\n\",\n  \"stderr\": \"\"\n}"}],"isError":false}}

2.3.Delete

DELETE /mcp HTTP/1.1
Host: mcp.litong.xyz
User-Agent: openai-mcp/1.0.0
Accept: application/json, text/event-stream
Accept-Encoding: gzip, br
Cdn-Loop: cloudflare; loops=1
Cf-Connecting-Ip: 57.154.187.41
Cf-Ipcountry: US
Cf-Ray: 987280c3bbc467be-SJC
Cf-Visitor: {"scheme":"https"}
Content-Type: application/json
Mcp-Protocol-Version: 2025-06-18
Mcp-Session-Id: c362f6dd171c4c0b9ee5d230ba011060
Traceparent: 00-68db9b0a00000000e8fb2ae346302db7-76b11a326432944c-00
Tracestate: dd=s:-1;p:11a91123b3c77b78;t.dm:-3
X-Datadog-Parent-Id: 1605897049399514698
X-Datadog-Sampling-Priority: -1
X-Datadog-Tags: _dd.p.tid=68db9b0a00000000,_dd.p.dm=-3
X-Datadog-Trace-Id: 16788059191577161143
X-Forwarded-For: 57.154.187.41, 172.69.22.186
X-Forwarded-Host: mcp.litong.xyz
X-Forwarded-Proto: http

HTTP/1.1 200 OK
date: Tue, 30 Sep 2025 08:55:45 GMT
server: t-io
content-type: application/json
mcp-session-id: c362f6dd171c4c0b9ee5d230ba011060
content-length: 0