xiaozhi-bridge
Connects a Xiaozhi AI voice agent to a Windows PC, enabling autonomous control via 58 MCP tools for screen vision, web browsing, file management, and system control.
README
🦞 PURPCLAW v7.0 — ULTIMATE
The AI Voice Agent MCP Bridge — Gives your Xiaozhi AI hardware full autonomous PC control via 58 MCP tools. Built by WEEMADSCOTSMAN at Pixel Dynasty.
🚀 New Here? Read the Full Setup Guide →
The setup guide covers everything from buying the hardware to your first voice command. Anyone with a Xiaozhi ball can use this.
What Is This?
PURPCLAW is an MCP (Model Context Protocol) bridge that connects a Xiaozhi AI voice agent to your Windows PC. It turns a voice-enabled AI ball into a fully autonomous engineering assistant that can:
- 🖥️ See your screen — Screenshots + OCR text recognition
- 📸 See YOU — Webcam capture, face/object detection
- 🌐 Browse the web — Playwright-powered: click links, fill forms, read pages
- 📁 Manage files — Read, write, copy, move, delete, search, zip/extract
- 🖱️ Control your desktop — Mouse clicks, keyboard typing, window management
- ⬇️ Download files — Direct URL downloads
- 📦 Install packages — pip, npm, or choco
- 🔊 Control audio — Volume up/down/mute/set
- 🔧 Run commands — Shell execution with injection protection
- 🧠 Remember things — Persistent memory across sessions
- 🦞 Run AI pipelines — Start/stop/monitor PURPCLAW build pipelines
Architecture
┌──────────────────┐ WebSocket ┌──────────────────┐
│ Xiaozhi Ball │ ◄──────────────── │ xiaozhi_bridge │
│ (DeepSeek V3.1) │ MCP Protocol │ v7.0.0 │
└──────────────────┘ └────────┬─────────┘
│
┌─────────────────┼─────────────────┐
│ │ │
┌─────┴─────┐ ┌──────┴──────┐ ┌─────┴─────┐
│ Playwright │ │ PowerShell │ │ Python │
│ Browser │ │ .NET APIs │ │ OpenCV │
└───────────┘ └─────────────┘ └───────────┘
58 MCP Tools
Vision (5)
| Tool | Description |
|---|---|
screen_capture |
Screenshot the screen |
screen_ocr |
Read text from screen via OCR |
screen_find_object |
Detect objects with YOLO |
screen_find_template |
Find image on screen (template match) |
screen_info |
Get monitor sizes |
Webcam (3)
| Tool | Description |
|---|---|
webcam_look |
Take a photo with PC webcam |
webcam_detect |
Detect people/faces/objects |
webcam_read |
OCR text visible to camera |
Mouse & Keyboard (3)
| Tool | Description |
|---|---|
mouse_click |
Click, double-click, right-click, drag |
mouse_scroll |
Scroll wheel |
keyboard_type |
Type text or press shortcuts |
UI Automation (1)
| Tool | Description |
|---|---|
find_and_click |
Find and click UI elements by text |
Window Management (3)
| Tool | Description |
|---|---|
window_list |
List open windows |
window_focus |
Focus a window by title |
window_close |
Close a window |
File System (8)
| Tool | Description |
|---|---|
file_read |
Read file contents |
file_write |
Write/append to files |
file_list |
List directory contents |
file_search |
Search by name or content (Node.js, fast) |
file_copy |
Copy files/directories |
file_move |
Move/rename files |
file_delete |
Delete files (system path protected) |
dir_create |
Create directories |
Browser (9) — Playwright-Powered
| Tool | Description |
|---|---|
browser_open |
Open URL with full interaction |
browser_click |
Click links/buttons by visible text |
browser_type |
Type into form fields |
browser_scroll |
Scroll pages up/down |
browser_get_content |
Read page text content |
browser_screenshot |
Screenshot the browser |
browser_navigate |
Navigate: goto/back/forward/reload |
browser_tabs |
List open tabs |
browser_close_tab |
Close a tab |
Downloads (1)
| Tool | Description |
|---|---|
download_file |
Download file from URL |
Process Management (2)
| Tool | Description |
|---|---|
process_list |
List running processes |
process_kill |
Kill process by name/PID (critical procs blocked) |
Audio (1)
| Tool | Description |
|---|---|
volume_control |
Set volume, mute/unmute, up/down |
Archives (2)
| Tool | Description |
|---|---|
zip_create |
Create zip archives |
zip_extract |
Extract zip archives |
Package Management (1)
| Tool | Description |
|---|---|
install_package |
Install via pip/npm/choco |
PURPCLAW Pipeline (4)
| Tool | Description |
|---|---|
purpclaw_start |
Start AI build pipeline |
purpclaw_stop |
Stop pipeline |
purpclaw_status |
Get pipeline status |
purpclaw_logs |
Read pipeline logs |
Git (1)
| Tool | Description |
|---|---|
git_command |
Run git commands |
Communication (4)
| Tool | Description |
|---|---|
http_request |
HTTP GET/POST/PUT/DELETE |
clipboard |
Read/write clipboard |
speak |
Text-to-speech via Kokoro TTS |
notification |
Desktop toast notifications |
Context & System (5)
| Tool | Description |
|---|---|
active_window |
Get focused window info |
system_status |
CPU, RAM, process overview |
disk_info |
Drive space for all disks |
network_info |
IP, WiFi, internet status |
execute_command |
Execute shell commands (injection-proof) |
Apps & Tasks (4)
| Tool | Description |
|---|---|
open_application |
Launch apps by name |
memory |
Persistent remember/recall/forget/list |
task_schedule |
Schedule delayed tasks |
task_list |
List scheduled tasks |
Setup
Prerequisites
- Node.js 18+
- Python 3.10+ with
opencv-python - Playwright (
npm i playwright) - Xiaozhi.me account + configured agent
Install
git clone https://github.com/weemadscotsman/purpclaw.git
cd purpclaw
npm install
cp .env.example .env
# Edit .env with your Xiaozhi MCP token
Run
# Windows (recommended)
start_xiaozhi_bridge.bat
# Or manually
set XIAOZHI_MCP_URL=wss://api.xiaozhi.me/mcp/?token=YOUR_TOKEN
node lib/xiaozhi_bridge.js
Environment Variables
| Variable | Description |
|---|---|
XIAOZHI_MCP_URL |
WebSocket URL from Xiaozhi MCP settings |
OPENCLAW_GATEWAY |
OpenClaw WebSocket (default: ws://127.0.0.1:18789) |
MCP_BRIDGE_URL |
Bridge HTTP server (default: http://localhost:3001) |
Security
- Shell injection protection — All user input is sanitized via
san()function - System path protection — Cannot delete Windows/Program Files/AppData
- Critical process protection — Cannot kill explorer, csrss, lsass, svchost
- Destructive command blocking — Blocks
format, recursive system deletes, shutdown - Tool timeout — Every tool has a timeout to prevent hanging
Tech Stack
- Runtime: Node.js (single-process, async event loop)
- Browser: Playwright (Chromium, headless: false)
- Vision: Python OpenCV + Haar cascade face detection
- Screen: .NET System.Drawing (fast screenshot via PowerShell)
- Shell: PowerShell -NoProfile -NonInteractive (fast startup)
- Protocol: MCP over WebSocket (JSON-RPC 2.0)
Built With
Created by Eddie Cannon (@weemadscotsman) — Edinburgh, Scotland 🏴
- Pixel Dynasty — Gaming Archive & Collab Hub
- NDK Threads — ndkthreads.com
- PixelBits420 — Fair-launch cryptocurrency
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。