MCP Feedback Tool

A Python implementation of a feedback collection tool similar to the mcp_claudeflow_collect_feedback tool, integrated with an MCP server for use with Claude and other MCP-compatible clients.

Features

• Collect user feedback via a customizable dialog window
• Store feedback history in a JSON file
• Format timestamps in various ways (full, ISO, date, time, unix)
• View feedback history
• Easily integrate with Claude Desktop or Claude Code via the MCP protocol
• Autoclose timer that automatically closes the dialog after 15 seconds (can be paused)
• Status messages for different interaction outcomes (normal, empty, cancelled, auto-closed, closed)

Installation

Prerequisites

• Python 3.8 or higher
• Tkinter (usually comes with Python)
• fastmcp package
• customtkinter package

Setup

1. Clone this repository or download the files
2. Install the required dependencies:

pip install -r requirements.txt

Usage

Standalone Feedback Collector

You can use the feedback collector directly without the MCP server:

from mcp_feedback import MCPFeedbackCollector

collector = MCPFeedbackCollector()

# Basic feedback collection
result = collector.collect_feedback()
print(f"Feedback received: {result['feedback']}")
print(f"Status: {result['status']}")
print(f"Timestamp: {result['timestamp']}")

# Custom prompt and title
result = collector.collect_feedback(
    prompt="What do you think about the new feature?",
    title="Feature Feedback",
    time_format="iso"
)

# View feedback history
collector.view_history()

Autoclose Timer

The feedback dialog includes an autoclose timer that will automatically close the dialog after 15 seconds of inactivity:

• A countdown timer shows the remaining time
• The "Pause" button allows users to pause/resume the timer
• If the dialog auto-closes, it returns a special message: "Session expired. The project is now in your hands. Proceed as you see fit."
• The timer resets to 15 seconds whenever the user:
  • Types in the feedback area
  • Moves the mouse within the dialog
  • Clicks anywhere in the dialog
  • Changes the appearance mode
  • Resumes the timer after pausing

Status Messages

The feedback collector provides different status messages based on how the dialog was closed:

Status	Description	Message
normal	User submitted feedback normally	User's feedback text
approved	User clicked Approve button	"I approve."
enough	User clicked Enough button	"That is good enough for now."
empty	User submitted with empty feedback	"Assume the user has no issues."
cancelled	User clicked Cancel	"Operation cancelled by user. Please stop all current operations."
auto_closed	Timer expired	"Assume the user has no issues."
closed	User closed window with X button	"Window closed by user. Please stop all current operations."

Note that:

• The auto_closed message is intentionally similar to the empty message to allow the AI to continue working on the project in both cases.
• Both cancelled and closed messages instruct the AI to stop all operations, providing consistent behavior when the user explicitly cancels or closes the dialog.
• The approved status indicates the user has approved the current state or suggestion without providing additional feedback.
• The enough status indicates the user finds the current state acceptable but may want to revisit it later.

MCP Server Integration

To use with Claude Desktop or Claude Code, you need to run the MCP server and configure your client to connect to it.

1. Start the MCP server:

python mcp_feedback_server.py

2. Configure Claude Desktop:

   • Open Claude Desktop settings
   • Navigate to the "Developer" tab
   • Click "Edit Config"
   • Add your MCP server configuration:

{
  "mcpServers": {
    "feedback": {
      "command": "python",
      "args": [
        "/absolute/path/to/mcp_feedback_server.py"
      ]
    }
  }
}

3. Configure Claude Code:

claude mcp add

Then provide the following details:

• Server Name: feedback
• Server Scope: Choose Project or Global
• Server Command: python /absolute/path/to/mcp_feedback_server.py

MCP Tools

The MCP server exposes two tools:

1. collect_feedback: Opens a dialog to collect user feedback

   • Parameters:
     • prompt: The message displayed to the user (default: "Please provide your feedback or describe your issue:")
     • title: The title of the feedback window (default: "AI Feedback Collection")
     • time_format: The format for time information (default: "full")
     • timezone: The timezone to use (default: local)

2. view_feedback_history: Returns a formatted string of all collected feedback

Customization

You can customize the appearance and behavior of the feedback collector by modifying the mcp_feedback.py file:

• Change the default window size
• Modify the UI styling
• Adjust the history file location
• Add more timestamp formats
• Change the autoclose timer duration

License

MIT

Acknowledgements

This project was inspired by the mcp_claudeflow_collect_feedback tool available in Claude.