This project provides an interactive feedback mechanism for AI development workflows, offering both a native desktop application and a web UI. It aims to improve efficiency and reduce costs by guiding AI to confirm operations with users, consolidating multiple tool calls into single, feedback-oriented requests. The target audience includes developers working in local, SSH remote, and WSL environments.

How It Works

The core concept is an "MCP" (likely Message, Command, or Process) server that facilitates communication between AI models and users. It uses a WebSocket connection for real-time feedback, allowing AI to adjust its behavior based on user input (text, image uploads, prompt selections). The dual interface architecture (Tauri-based desktop app and a web UI) ensures flexibility across different deployment scenarios, with both interfaces offering unified functionality.

Quick Start & Requirements

• Install uv: pip install uv
• Configure MCP in mcp-config.json (examples provided for web and desktop modes).
• AI assistant requires prompt engineering rules: MCP Interactive Feedback Rules follow mcp-feedback-enhanced instructions
• Official documentation and examples are available within the repository.

Highlighted Details

• Cross-platform desktop application support via Tauri (Windows, macOS, Linux).
• Web UI for remote and WSL environments.
• Prompt management with CRUD operations and usage statistics.
• Auto-timed submission and session tracking with local file storage.
• Markdown rendering for AI work summaries.
• Full image format support (PNG, JPG, GIF, BMP, WebP) with drag-and-drop and clipboard paste.
• SSH remote development support with MCP_WEB_HOST configuration.

Maintenance & Community

• Active development with recent updates addressing SSH support, macOS compilation, and packaging.
• Community support via Discord: https://discord.gg/Gur2V67.
• Key contributors include Fábio Ferreira (original author), penn201500, leo108, Alsan, and fireinice.

Licensing & Compatibility

• MIT License.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

AI models may exhibit instability in image parsing, requiring retries or adjustments to image quality/format. The project previously supported a PyQt6 GUI, but this was removed in v2.4.0; legacy versions are available for users needing that specific interface. UV cache can accumulate significant disk space and requires periodic cleanup.