Leo Feedback MCP

A customized MCP (Model Context Protocol) server for interactive feedback collection during AI-assisted development. Built with a Flutter Web frontend and Python/FastAPI backend.

Features

• Flutter Web UI with dark theme for interactive feedback
• AI Work Summary panel with Markdown rendering
• Image support - upload, drag & drop from IDE, paste (PNG, JPG, GIF, BMP, WebP)
• Auto-submit countdown with configurable prompts
• Audio & browser notifications when AI requests feedback
• Session history - persists across browser reloads via backend storage
• Drag & drop files from IDE (Cursor) into the feedback panel
• Keyboard shortcuts - Ctrl+Enter / Cmd+Enter for quick submit
• Smart browser detection (WSL, SSH, local)
• Session management with auto-cleanup

Architecture

┌─────────────────────────────────────────────┐
│  Cursor / AI Client                         │
│  (calls leo_feedback_mcp MCP tool)           │
└──────────────┬──────────────────────────────┘
               │ MCP Protocol (stdio)
┌──────────────▼──────────────────────────────┐
│  Python Backend                             │
│  ├── FastMCP server (MCP tool definitions)  │
│  ├── FastAPI (HTTP routes + WebSocket)      │
│  └── Session management & image processing  │
└──────────────┬──────────────────────────────┘
               │ HTTP + WebSocket
┌──────────────▼──────────────────────────────┐
│  Flutter Web Frontend                       │
│  ├── Workspace (AI Summary + Feedback)      │
│  ├── Sessions (chat history)                │
│  ├── Settings (auto-submit, timeout)        │
│  └── About                                  │
└─────────────────────────────────────────────┘

Requirements

• Python 3.11+
• uv (Python package manager)
• Flutter 3.x+ (for frontend development/building)

Installation

Option A: Install from PyPI (recommended for users)

pip install leo-feedback-mcp

Or with uv:

uv pip install leo-feedback-mcp

MCP configuration (.cursor/mcp.json):

{
  "mcpServers": {
    "leo-feedback-mcp": {
      "command": "uvx",
      "args": ["leo-feedback-mcp"],
      "timeout": 180,
      "autoApprove": ["leo_feedback_mcp"]
    }
  }
}

Option B: Install from source (for development)

git clone <your-repo-url> ~/Desktop/leo-feedback-mcp
cd ~/Desktop/leo-feedback-mcp
make init
make build

MCP configuration (.cursor/mcp.json):

{
  "mcpServers": {
    "leo-feedback-mcp": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "<path-to-leo-feedback-mcp>",
        "leo-feedback-mcp"
      ],
      "timeout": 180,
      "autoApprove": ["leo_feedback_mcp"]
    }
  }
}

Replace <path-to-leo-feedback-mcp> with the actual path to the cloned repository.

Apply configuration

You can place the MCP config in:

• Project-level: .cursor/mcp.json in your project root
• Global: ~/.cursor/mcp.json to apply across all projects

After editing, restart Cursor or reload MCP servers (Cursor Settings > MCP > Reload).

Usage

Once configured, the AI assistant will automatically call leo_feedback_mcp to collect your feedback during conversations. A Web UI will open in your browser where you can:

1. Review the AI's work summary (Markdown rendered)
2. Provide text feedback with keyboard shortcuts
3. Attach images via upload button, drag & drop, or paste
4. View session history of previous AI interactions
5. Configure auto-submit prompts and timeout settings

Environment Variables

Credits

• Author: Leo Nguyen