中文版

一款用于光标IDE的MCP（模型上下文协议）工具，允许AI在单次请求内暂停等待用户反馈，通过将多回合交互保持在同一会话中来降低计费成本。

为什么？

光标按请求计费（每月有限）。没有这个工具，每轮反馈都会支付一个新的请求费用。使用交互反馈MCP，AI会通过弹窗暂停等待你的输入，然后继续——所有操作都在一个请求内完成。这之所以有效，是因为请求内的工具调用是免费的。

特色

• 单请求反馈循环——AI暂停并等待用户输入，不消耗额外请求
• 始终在顶部的原生Win32反馈窗口
• 将文件拖放到窗口或粘贴复制文件以插入路径
• 无人值守操作（单次和循环模式）的自动回复规则
• 窗口管理——当不再需要时，空窗口关闭;有用户输入的窗口保持开启，以避免文本丢失
• 所有以时间戳记录为[[T0]的交互（AI请求/用户回复/自动回复）

平台支持

Interactive Feedback MCP

中文版

An MCP (Model Context Protocol) tool for Cursor IDE that lets the AI pause and wait for user feedback within a single request, reducing billing costs by keeping multi-turn interactions in one session.

Why?

Cursor bills by request (limited monthly). Without this tool, each round of feedback costs a new request. With Interactive Feedback MCP, the AI pauses and waits for your input via a popup window, then continues — all within one request. This works because tool calls within a request are free.

Features

• Single-request feedback loop — AI pauses and waits for user input without consuming extra requests
• Always-on-top native Win32 feedback window
• Drag & drop files onto the window or paste copied files to insert their paths
• Auto-reply rules for unattended operation (oneshot and loop modes)
• Window management — empty windows close when no longer needed; windows with user input stay open to avoid losing text
• All interactions (AI request / user reply / auto-reply) logged with timestamps to feedback_log.txt

Platform Support

Quick Start (Windows)

1. Download

Download the latest release from Releases. You’ll get three files:

• feedback-server.exe — the MCP server
• feedback-gui.exe — the GUI window
• feedback.exe — command-line version (optional)

Place them in the same folder (e.g. C:\interactive-feedback-mcp\).

2. Configure MCP

Open Cursor Settings → MCP → Add a new MCP server, or manually edit your mcp.json:

For global use — edit %USERPROFILE%\.cursor\mcp.json:

{
  "mcpServers": {
    "interactive-feedback-mcp": {
      "command": "C:\\path\\to\\feedback-server.exe",
      "timeout": 600,
      "autoApprove": ["interactive_feedback"]
    }
  }
}

For a specific project — create .cursor/mcp.json in the project root with the same content.

Replace the path with the actual location on your machine. Use double backslashes in JSON. timeout is in seconds — 600 gives you 10 minutes to respond. autoApprove skips the manual approval popup each time the AI calls the tool.

3. Configure Cursor Rules

Add the following to Cursor Settings → Rules:

After completing your full response, you MUST call interactive-feedback-mcp as your very last action. Do NOT call it in the middle of your answer.
After receiving the MCP result, if there is feedback, process it first. Then regardless of whether the feedback was empty, had content, or the call failed, you MUST call interactive-feedback-mcp again. NEVER stop. Loop forever.