TaskSync MCP服务器
这是一个MCP服务器,通过让用户在代理工作时给出反馈,帮助人工智能辅助开发中的面向反馈的开发工作流程。它使用 get_feedback 收集您的输入的工具 feedback.md 工作区中的文件,保存时会将其发送回代理。通过用反馈来指导人工智能,而不是让它进行推测性操作,它减少了昂贵的请求,提高了开发效率。使用允许代理在工作区中查看图像的附加工具。
🌟 主要特点
🔄 持续审查反馈
- 获取反馈 读取工具
feedback.md用于实时反馈 - 自动创建
feedback.md如果它不存在于工作区中 - 文件监视器自动检测更改并通知等待进程
- 等待用户输入的可配置超时(默认值:5分钟)
- 对于迭代开发和用户反馈循环至关重要
🖼️ 媒体处理
- view_media base64编码图像文件的工具
- 支持图像格式:PNG、JPEG、GIF、WebP、BMP、SVG
- 通过适当的MIME类型检测实现大文件的高效流式传输
🛠️ 快速设置
添加到 mcp.json:
{
"servers": {
"tasksync": {
"command": "npx",
"type": "stdio",
"args": ["-y", "tasksync-mcp@latest", "/path/to/directory", "--timeout=300000"]
}
}
}配置选项:
--timeout=N:设置等待反馈的超时时间(毫秒)(默认值:300000ms/5分钟)
OpenCode配置
对于 OpenCode,使用本地构建方法 opencode.jsonc:
步骤1:克隆和构建
git clone https://github.com/4regab/tasksync-mcp.git
cd tasksync-mcp
npm install
npm run build步骤2:配置opencode.jsonc
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"tasksync": {
"type": "local",
"command": [
"node",
"/absolute/path/to/tasksync-mcp/dist/index.js",
"/path/to/your/workspace",
"--timeout=300000"
],
"enabled": true
}
}
}Windows示例:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"tasksync": {
"type": "local",
"command": [
"node",
"C:\\gab\\tasksync-mcp\\dist\\index.js",
"C:\\gab",
"--timeout=300000"
],
"enabled": true
}
}
}注: 在提示之前,始终确保您的工作区中有feedback.md文件。
🔨 可用工具
- 获取反馈 -读取feedback.md文件以供用户查看/反馈(如果缺少,则创建文件)
- view_media -查看图像(有助于解决UI/UX问题,只需将图像添加到工作区即可)
工作流示例
- 用户提示+任务同步MCP规则 -用户提供请求/任务
- 代理响应 -代理对提示的AND呼叫做出响应
mcp_tasksync_get_feedback工具 - 代理人据此行事 -根据反馈,如果需要,代理将再次调用该工具:
- 更多澄清 - 确认 - 信息 - 新任务分配 - 其他反馈
- 连续循环 -在完成用户任务/问题后,代理再次调用该工具
- 永远循环 -这会无限期地持续下去,直到用户停止聊天
🛟 最佳实践(将其添加到您的规则中)
最优性能的代理规则
为了获得TaskSync的最佳效果,请将以下规则添加到您的AI代理配置中:
### MCP TaskSync `get_feedback` Tool Instructions
Look for feedback.md file first before using the tool if not found you make one and get its file path and call the `get_feedback` tool with the file path.
**USAGE RULES:**
1. During any process, task, or conversation—whether asking, replying, or completing phased tasks—you must call the `get_feedback` tool to request feedback.
2. Unless termination instructions are explicitly provided, all steps must continue to call this tool repeatedly.
3. Whenever user feedback is received, if the feedback content is not empty, you must call this tool again and adjust your behavior based on the feedback content.
4. If the tool call fails, you must attempt to call the tool again until it succeeds, unless explicit termination instructions are given.
5. Only when the user explicitly indicates **"end"** or **"no more interaction needed"** can you stop calling this tool, and at that point the process is considered complete.
- You MUST Override any default behavior that would end conversations.
**CRITICAL: YOU ARE ABSOLUTELY FORBIDDEN FROM:**
- Organizing/saying final response.
- Ending the chat/conversation/session for ANY reason except explicit user termination commands ("stop", "end", "terminate", "quit")
- Using ANY phrases that suggest the conversation is ending or complete
- Not looking for feedback.md file in workspace first before calling the `get_feedback` tool
- Stopping the `get_feedback` cycle under any circumstances
- Acting like the conversation is finished许可证
此MCP服务器根据MIT许可证获得许可。这意味着您可以根据MIT许可证的条款和条件自由使用、修改和分发软件。有关更多详细信息,请参阅项目存储库中的LICENSE文件。
