Skip to content
| Marketplace
Sign in
Visual Studio Code>Programming Languages>DevLinker - MCP HostNew to Visual Studio Code? Get it now.
DevLinker - MCP Host

DevLinker - MCP Host

sanchaii

|
1,446 installs
| (0) | Free
A Copilot chat participant that connects your AI assistants with your development environment.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

DevLinker - MCP Host 插件

GitHub VSCode Extension

使用方法

  1. 打开VS Code中的Copilot聊天窗口

  2. 使用扩展聊天角色@devlinker

    @devlinker

  3. 通过 /connectStdio 指令以Stdio的方式连接MCP服务,输入部分为执行命令及参数,如果文件路径中有空格,请使用双引号

    @devlinker /connectStdio node D://xxxx/xxx/xx.js

    @devlinker /connectStdio python D://xxxx/xxx/xx.py

    @devlinker /connectStdio python "D://xxxx/xx xx/xx.py"

  4. 或者通过 /connectSSE 指令以SSE的方式连接MCP服务

    @devlinker /connectSSE http://localhost:8000/sse

  5. 或者通过 /connectSHttp 指令以Streamable Http的方式连接MCP服务,如果需要添加自定义请求头,请使用 /load 指令配合json文件来连接

    @devlinker /connectSHttp https://example.com/mcp

  6. 连接成功后会有Connection id返回,事后可以使用此id来断开连接

  7. LLM模型建议GPT-4o或Claude 3.5 Sonnet,其它模型支持力度不一

特性

  1. 支持服务端特性:Tools、Resources(在不使用/autoContext指令的前提下将进行弹窗,需用户自己选择引用的Resources),并监听相应列表更新事件
  2. 支持客户端特性:Roots以及Sampling
  3. 支持智能跟进建议生成
  4. 支持管理多个MCP服务连接
  5. 支持主动断开连接、主动重新连接MCP服务等功能
  6. 支持从本地json文件导入并连接MCP服务,须遵循如下格式:
    {
        "Stdio": [
            "node D:/path/to/your/jsfile.js",
            "node D:/path/to/your/jsfile.js D:/path/to/target/folder",
            "python E:/path/to/your/pyfile.py"
        ],
        "SSE": [
            "http://xxx.xx.xx.x:8000/sse",
            "https://xxxxxx.com/sse"
        ],
        "StreamableHttp": [
            {
                "url": "https://example1.com/mcp"
            },
            {
                "url": "https://example2.com/mcp",
                "requestHeaders": {
                    "Authorization": "Bearer token123",
                    "Content-Type": "application/json"
                }
            }
        ]
    }

常用聊天指令

  • /connectStdio - 以Stdio的方式连接MCP服务

  • /connectSSE - 以SSE的方式连接MCP服务

  • /connectSHttp - 以Streamable Http的方式连接MCP服务

  • /disconnect - 断开某个MCP连接,后续接connectionID使用,连接成功后会返回id

    @devlinker /disconnect connectionID

  • /disconnectAll - 断开所有MCP连接

  • /refresh - 刷新所有MCP服务连接

  • /load - 将MCP服务json配置文件拖拽至聊天窗口作为附件后,通过此指令进行全部加载并连接

  • /autoContext - 由LLM选择引用哪些外部资源,但是要慎用,因为存在LLM使用旧数据的隐患,需要用户显式的说明重新拉取,才会获得最新数据

常见问题

  • 在创建并连接本地MCP服务进程前建议先通过本地终端执行,排查问题

  • 在显示已连接MCP服务为0时,窗口内依然存在断开连接等followups时不用在意,再次对话将会刷新

  • 如果每次对话都返回同个错误,解决不了时尝试新建一个聊天

DevLinker - MCP Host Plugin

How to Use

  1. Open the Copilot chat window in VS Code

  2. Use the extension chat role @devlinker

    @devlinker

  3. Connect to MCP service by using /connectStdio command. If there are spaces in file paths, use double quotes

    @devlinker /connectStdio node D://xxxx/xxx/xx.js

    @devlinker /connectStdio python D://xxxx/xxx/xx.py

    @devlinker /connectStdio python "D://xxxx/xx xx/xx.py"

  4. Connect to MCP service by using /connectSSE command.

    @devlinker /connectSSE http://localhost:8000/sse

  5. Connect to MCP service by using /connectSHttp command. If you need to add custom request headers, please use the /load command with a json file to connect.

    @devlinker /connectSHttp https://example.com/mcp

  6. After successful connection, a Connection id will be returned, which can be used later to disconnect

  7. Recommended LLM model are GPT-4o, Claude 3.5 Sonnet, other models may have limited support

Features

  1. Supports MCP server features: Tools, Resources (without using the /autoContext command, a popup will appear requiring users to select which Resources to reference), and listening for respective list update events

  2. Supports MCP client features: Roots and Sampling

  3. Supports smart followups suggestions

  4. Supports managing multiple MCP service connections

  5. Supports actively disconnecting and reconnecting MCP services

  6. Supports importing and connecting MCP services from local json files, which must follow this format:

    {
        "Stdio": [
            "node D:/path/to/your/jsfile.js",
            "node D:/path/to/your/jsfile.js D:/path/to/target/folder",
            "python E:/path/to/your/pyfile.py"
        ],
        "SSE": [
            "http://xxx.xx.xx.x:8000/sse"
        ],
        "StreamableHttp": [
            {
                "url": "https://example1.com/mcp"
            },
            {
                "url": "https://example2.com/mcp",
                "requestHeaders": {
                    "Authorization": "Bearer token123",
                    "Content-Type": "application/json"
                }
            }
        ]
    }

Common Chat Commands

  • /connectStdio - Connect to MCP service using Stdio (Standard Input/Output)

  • /connectSSE - Connect to MCP service using SSE (Server-Sent Events)

  • /connectSHttp - Connect to MCP service using Streamable HTTP

  • /disconnect - Disconnect a specific MCP connection, use the connectionID that was returned after successful connection

    @devlinker /disconnect connectionID

  • /disconnectAll - Disconnect all MCP connections

  • /refresh - Refresh all MCP service connections

  • /load - After dragging an MCP service configuration json file into the chat window as an attachment, use this command to load and connect to all services

  • /autoContext - Let the LLM choose which external resources to reference. Use with caution as there is a risk of the LLM using outdated data. Users need to explicitly request refreshing data to get the latest information

Common Issues

  • Before creating and connecting to a local MCP service process, it is recommended to first execute it through your local terminal to troubleshoot any issues

  • When the display shows 0 connected MCP services but disconnect and other followups still appear in the chat window, don't worry about it. The next conversation turn will refresh these options

  • If every conversation returns the same error and you can't resolve it, try starting a new chat

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2025 Microsoft