English | 简体中文
MCP server for collecting interactive user feedback during AI-assisted development. By guiding AI to confirm with users instead of guessing, it reduces unnecessary tool calls, cuts platform costs, and improves development efficiency.
- AI calls the
interactive_feedbacktool - Browser interface (Web UI) opens automatically
- User provides feedback text, uploads screenshots, or picks a saved prompt
- Feedback is delivered to AI in real-time via WebSocket
- AI adjusts its behavior or completes the task based on feedback
pip install uvAdd the following to your MCP configuration file:
{
"mcpServers": {
"mcp-interactive": {
"command": "uvx",
"args": ["mcp-interactive@latest"],
"timeout": 600,
"autoApprove": ["interactive_feedback"]
}
}
}Supported AI platforms: Cursor | Cline | Windsurf | Augment | Trae
| Variable | Description | Default |
|---|---|---|
MCP_WEB_HOST |
Web UI bind address | 127.0.0.1 |
MCP_WEB_PORT |
Web UI port | 9766 |
MCP_LANGUAGE |
UI language (zh-CN / en) |
Auto-detect |
MCP_DEBUG |
Debug mode | false |
For SSH remote development, set MCP_WEB_HOST to 0.0.0.0 to allow remote access, or use SSH port forwarding.
- Prompt management - CRUD for saved prompts with smart sorting
- Auto-timed submit - Configurable timer (1-86400s) with pause/resume
- Image upload - Drag & drop, paste from clipboard; PNG/JPG/GIF/BMP/WebP
- Multi-language - Simplified Chinese, English, instant switching
- WebSocket real-time - Status monitoring, auto-reconnect
git clone https://github.com/RealAlexandreAI/mcp-interactive.git
cd mcp-interactive
uv sync
# Test
make test # Unit tests
make test-web # Web UI test
# Code quality
make check # Full check (lint + format + type)- Original project: noopstudios/interactive-feedback-mcp by Fabio Ferreira
- Upstream fork: Minidoracat/mcp-feedback-enhanced
MIT
