mcp-feedback-collector

mcp-feedback-collector is an MCP server designed to collect interactive user feedback during AI-assisted workflows. It exposes a streamlined collect_feedback tool that lets the AI report progress and the user provide textual and/or image feedback. The server also supports auxiliary utilities like pick_image for selecting a single image and get_image_info to inspect image metadata. By integrating with your MCP-enabled assistant, this server helps ensure user feedback is captured before completing a task, enabling iterative improvement and reliable handoffs.

To use it, configure your MCP host (for example within Claude Desktop or another MCP client) to launch the mcp-feedback-collector via uvx. The provided configuration sets environment variables to control encoding and dialog timeout. Once configured and started, the AI can call collect_feedback with a status message, optionally attach images, and the user can respond with text and visual feedback. When feedback is exhausted, the session ends automatically, ensuring a clean, user-driven loop for feedback collection.

Prerequisites:

• Python 3.8 or newer
• uvx package (installable via pip)
• Access to an MCP environment where the server will be registered

Step-by-step:

1. Install uvx if you don’t have it:

pip install uvx

2. Install or prepare the mcp-feedback-collector package via uvx. This typically involves enabling the MCP server package for the collector:

uvx install mcp-feedback-collector

3. Create or update your MCP configuration to register the server. For example, in your MCP manager configuration (config.json or equivalent), add:

{
  "mcpServers": {
    "mcp-feedback-collector": {
      "command": "uvx",
      "args": ["mcp-feedback-collector"],
      "env": {
        "PYTHONIOENCODING": "utf-8",
        "MCP_DIALOG_TIMEOUT": "600"
      }
    }
  }
}

4. Restart your MCP host/application to apply the new server configuration.
5. Verify the server is reachable by invoking its collect_feedback tool through the MCP interface, then begin providing feedback via the UI or API.

Environment variables:

• MCP_DIALOG_TIMEOUT controls how long the dialog waits for user input before timing out (default 300 seconds; recommended 600 for longer tasks).
• PYTHONIOENCODING ensures consistent text encoding when handling user input and feedback. Common issues:
• If uvx cannot locate the mcp-feedback-collector package, ensure the package name is correct in your uvx install command and that your environment has network access to fetch it.
• Ensure the MCP client supports invoking the collect_feedback, pick_image, and get_image_info tools as demonstrated in the README example.
• When running in headless environments, verify that image handling and any GUI components are either disabled or properly mocked, since the core feedback collection relies on textual input and optional image data. Configuration tips:
• Adjust MCP_DIALOG_TIMEOUT according to the expected length of user interactions.
• Add additional environment variables if your deployment requires custom encodings, proxies, or logging settings.