MCP Feedback Collector

中文版本 | English Version

A feedback collection tool that supports text and image feedback with a dark theme design.

🌐 Web Version Available

The MCP Feedback Collector (Web Version) is specifically designed for development scenarios where you connect to remote servers via SSH. The Web version solves the limitations of traditional local GUI feedback collection methods in remote development environments.

🔄 Project Relationship

• This Project: Traditional local GUI feedback collection tool
• Web Version: Designed specifically for remote development environments
• Choose the right tool based on your needs:
  • If you develop locally and prefer native GUI, use this project
  • If you develop on remote servers via SSH or prefer web interfaces, use the web version

🎯 Problems Solved

• Remote Development Limitations: SSH connections cannot use local GUI interfaces
• Server Environment Compatibility: Perfect for headless server environments
• Web-based Interface: Access feedback collection through any web browser

👉 Visit Web Version →

Interface Preview

Chinese Interface	English Interface

Why Do You Need This Tool?

When using Cursor, you may have noticed that AI assistants often stop after completing a single task, even though a single request can still use many more tool calls. This leads to resource waste, as each request can theoretically call up to 25 tools.

The core purpose of this MCP tool is:

• 🔄 Continue Conversations: In a single fast Request, ensure the AI actively calls this MCP tool to collect feedback after completing tasks (instead of simply terminating the conversation), allowing for subsequent tasks
• 🔥 Maximize Utilization: Let each fast Request fully utilize the 25 tool calls quota
• 💰 Optimize Costs: Maximize the value return of each fast Request

Before vs After Comparison

	Before	After
Workflow	Input in Agent mode: Complete <Task 1> ⬇ Task completed, this fast request ends	Input in Agent mode: Complete <Task 1> ⬇ Task completed, call feedback collector for feedback ⬇ Input in MCP tool: Complete <Task 2> ⬇ Task completed, call feedback collector for feedback ⬇ Input in MCP tool: Complete <Task 3> ⬇ Task completed, call feedback collector for feedback ⬇ ...... ⬇ until reaching 25 tool call limit
Request Consumption	Consumes 1 Request	Consumes 1 Request

Features

• 🎨 Dark Theme Interface: Dark color scheme consistent with Cursor editor
• 📝 Multi-line Text Feedback: Support for multi-line text input and editing
• 🖼️ Image Feedback Support: Support for image selection and clipboard pasting, including image thumbnail display and deletion
• 🌍 Chinese/English Interface Toggle: Support for interface language switching
• ⌨️ Keyboard Shortcuts: Quick submit (⌘+Enter/Ctrl+Enter) and cancel (ESC) operations

System Requirements

• Python 3.8+
• tkinter (usually installed with Python)
• Environment with GUI support

Installation

1. Clone Repository

git clone <repository-url>
cd feedback_collector

2. Create Virtual Environment

conda create -n feedback_collector python=3.11
conda activate feedback_collector

3. Install Dependencies

pip install -r requirements.txt

MCP Configuration

Add the following configuration to Cursor's MCP configuration file:

macOS/Linux Configuration

{
  "mcpServers": {
    "feedback-collector": {
      "command": "/path/to/your/python",
      "args": [
        "/path/to/feedback_collector/src/server.py"
      ],
      "env": {
        "MCP_DIALOG_TIMEOUT": "600",
        "LANGUAGE": "EN"
      }
    }
  }
}

Example configuration file: mcp_config_example_for_mac_and_linux.json

Windows Configuration

Method 1: Direct Python execution

{
  "mcpServers": {
    "feedback-collector": {
      "command": "/path/to/your/conda/envs/feedback_collector/python.exe",
      "args": [
        "/path/to/feedback_collector/src/server.py"
      ],
      "env": {
        "MCP_DIALOG_TIMEOUT": "600",
        "LANGUAGE": "EN"
      }
    }
  }
}

Example configuration file: mcp_config_example_for_windows_1.json

Method 2: Using cmd wrapper

{
  "mcpServers": {
    "feedback-collector": {
      "command": "cmd",
      "args": [
        "/c",
        "/path/to/your/conda/envs/feedback_collector/python.exe",
        "/path/to/feedback_collector/src/server.py"
      ],
      "env": {
        "MCP_DIALOG_TIMEOUT": "600",
        "LANGUAGE": "EN"
      }
    }
  }
}

Example configuration file: mcp_config_example_for_windows_2.json

Configuration Description:

• command: Python interpreter path or system command
• args: Arguments passed to the command
• MCP_DIALOG_TIMEOUT: Dialog timeout in seconds, default 600 seconds
• LANGUAGE: Interface language, CN (Chinese) or EN (English)

!MCP Server Configuration

Getting Paths

macOS/Linux:

# Get Python path
conda activate feedback_collector
which python

# Get project path
cd /path/to/feedback_collector
pwd

Windows:

# Get Python path
conda activate feedback_collector
python -c "import sys; print(sys.executable)"

Project Rules Configuration

Add the following rules to Cursor's Project Rules (set Rule Type to Always) to ensure automatic feedback collection after each task completion:

Any time you need to ask the user a question, make sure to use the <feedback-collector.collect_feedback> MCP.

After completing each user-specified task, the <feedback-collector.collect_feedback> mcp tool must be used to collect user feedback.

Before finishing a user Request, make sure to call the <feedback-collector.collect_feedback> MCP rather than just ending the task.

Keep calling <feedback-collector.collect_feedback> MCP until the user's feedback is empty, then end the Request.

If the ongoing task involves UI modifications, you should only complete the user's coding task and must not perform testing yourself. The correct approach is to prompt the user to test it themselves, and then use the <feedback-collector.collect_feedback> MCP tool to obtain their feedback.

Configuration Steps:

1. Open Cursor settings
2. Find Project Rules option
3. Add new rule and paste the above content
4. Set Rule Type to Always
5. Save configuration

!Rules Configuration

MCP Tools

collect_feedback()

Main tool for collecting user feedback:

• Display feedback collection interface
• Support text and image feedback
• Return structured feedback data

Usage

1. Complete the above MCP configuration and Project Rules configuration
2. In conversations with AI, the AI will automatically call the feedback collector after completing current tasks (instead of simply terminating the conversation)
3. Enter text feedback or add images in the popup dialog
4. Use shortcut ⌘+Enter (Mac) or Ctrl+Enter (Windows/Linux) to submit feedback; press ESC to cancel
5. AI will continue subsequent operations based on feedback (to achieve full utilization of up to 25 tool calls per single request)

Important: Enable Cursor Settings -> Features -> Chat -> Enable auto-run mode for the smoothest user experience, but make sure to configure command allowlist/denylist properly for security

Project Structure

feedback_collector/
├── README.md                 # Project documentation (English)
├── README_CN.md              # Project documentation (Chinese)
├── requirements.txt          # Dependencies list
└── src/                     # Source code directory
    ├── server.py            # MCP server entry point
    ├── ui/                  # UI modules
    │   ├── theme.py         # Dark theme management
    │   ├── components.py    # macOS style components
    │   └── feedback_dialog.py # Feedback collection dialog
    ├── core/                # Core business logic
    │   ├── feedback_collector.py # Feedback collector
    │   └── image_handler.py # Image processing
    └── utils/               # Utility functions
        ├── gui_utils.py     # GUI utility functions
        └── i18n.py          # Internationalization support