yyds-auto（元达汽车）

Yyds.Auto — Android RPA Skill for AI Agents

Let LLMs directly control Android devices through the MCP protocol.

Yyds.Auto is a production-grade Android RPA (Robotic Process Automation) platform that exposes 60 MCP tools covering the full spectrum of Android device automation — from pixel-level touch injection and OCR to UI hierarchy inspection, file management, and on-device AI agent orchestration.

What Can It Do?

Architecture

AI Agent (Claude / GPT / Gemini / Cursor / Windsurf / ...)
  ↓ MCP Protocol (stdio, JSON-RPC)
yyds-auto-mcp (Node.js, this skill)
  ↓ HTTP REST (JSON, port 61140)
yyds.py engine (Android, aiohttp server)
  ↓ IPC
yyds.auto engine (Android, kernel-level UI automation)

The MCP server communicates with the on-device engine via HTTP REST. When connected via USB, ADB port forwarding is set up automatically. Remote devices over WiFi/LAN are also supported.

Prerequisites

1. Android device with Yyds.Auto installed and the engine running
2. Connection: USB (auto ADB forward) or WiFi (same LAN)
3. Node.js >= 18

Quick Start

Install the MCP Server

npm install -g yyds-auto-mcp

Connect to a USB Device (auto-detected)

# Default: 127.0.0.1:61140, ADB forward set up automatically
yyds-auto-mcp

Connect to a Remote Device

YYDS_DEVICE_HOST=192.168.1.100 YYDS_DEVICE_PORT=61140 yyds-auto-mcp

Claude Desktop Configuration

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "yyds-auto": {
      "command": "npx",
      "args": ["-y", "yyds-auto-mcp"],
      "env": {
        "YYDS_DEVICE_HOST": "127.0.0.1",
        "YYDS_DEVICE_PORT": "61140"
      }
    }
  }
}

Cursor / Windsurf / VS Code Configuration

Add the same MCP server configuration in your editor's MCP settings.

Environment Variables

Tool Reference

Device Information

• device_info — Comprehensive device info: engine version, screen size, IMEI, foreground app
• get_foreground_app — Current foreground app & Activity
• get_screen_size — Device screen resolution
• is_network_online — Check network connectivity

Touch & Input

• tap *(x, y, count?, interval?)* — Tap at coordinates (supports multi-tap)
• swipe *(x1, y1, x2, y2, duration?)* — Swipe gesture
• long_press *(x, y, duration?)* — Long press at coordinates
• drag *(x1, y1, x2, y2, duration?)* — Drag from point A to B
• input_text *(text)* — Input text into the focused field
• set_clipboard / get_clipboard — Clipboard operations
• press_key *(key)* — Press a key (home, back, enter, etc.)

Screenshot

• take_screenshot *(quality?)* — Returns base64 JPEG image (LLM directly interprets it)
• save_screenshot *(path?)* — Save screenshot to device storage

UI Automation

• dump_ui_hierarchy — Full UI tree (auto-trimmed when >15KB to save tokens)
• find_ui_elements *(text?, resourceId?, className?, clickable?, ...)* — Find elements by attributes
• get_element_relation *(hashcode, type?)* — Get parent/children/sibling of an element
• wait_for_element *(text?, resourceId?, timeout?)* — Wait until an element appears
• scroll_to_find *(text?, direction?, maxScrolls?)* — Scroll until an element is found

OCR & Image

• screen_ocr *(x?, y?, w?, h?)* — Recognize text on screen (region supported)
• tap_text *(text, index?)* — OCR + tap on the matching text
• image_ocr *(path)* — Recognize text from an image file
• find_image_on_screen *(templates, threshold?)* — Template matching
• get_pixel_color *(x, y)* — Get pixel color at coordinates
• compare_images *(image1, image2)* — Image similarity comparison
• wait_for_screen_change *(timeout?, threshold?)* — Wait for the screen to change

Shell

• run_shell *(command)* — Execute shell commands with elevated privileges

App Management

• launch_app / stop_app *(packageName)* — Start/stop apps
• list_installed_apps — List all non-system installed apps
• is_app_running *(packageName)* — Check if an app is running
• open_url *(url)* — Open URL in browser
• show_toast *(message)* — Display a toast notification
• install_apk / uninstall_app — Install/uninstall apps

File Operations

• list_files / read_file / write_file — Browse, read, write files
• file_exists / delete_file / rename_file / create_directory

Script Projects

• list_projects / project_status — View Python projects
• start_project / stop_project — Control project execution
• run_python_code *(code)* — Execute Python code snippets on the device

Pip Management

• pip_list / pip_install / pip_uninstall / pip_show

AI Agent

• agent_run *(instruction)* — Run an on-device AI agent with natural language
• agent_stop / agent_status — Control and monitor the agent
• agent_get_config / agent_set_config — Configure AI provider & model
• agent_get_providers / agent_get_models — List available providers & models
• agent_test_connection — Verify AI model connectivity

Key Features

🔄 Auto-Reconnect

USB connection drops are handled gracefully — when the device disconnects, the MCP server automatically re-establishes ADB port forwarding and retries the request.

🚀 Auto-Bootstrap

On first connection via USB, the server automatically sets up ADB forwarding and starts the engine on the device if it's not already running.

🧠 Smart UI Dump

UI hierarchy dumps over 15KB are automatically trimmed to keep only actionable elements (those with text, resource-id, content-desc, or clickable/scrollable attributes), reducing LLM token usage.

🎯 Kernel-Level Touch

Touch events are injected at the Linux kernel level, making them work in any app including games, locked-down apps, and areas that block accessibility-based input.

Example Prompts

Once connected, try these prompts with your AI agent:

• *"Take a screenshot and describe what's on the screen"*
• *"Open WeChat and send 'Hello' to the first chat"*
• *"Find all buttons on the screen and list their labels"*
• *"OCR the screen and find any phone numbers"*
• *"Swipe up 3 times to scroll through the feed"*
• *"Install the APK at /sdcard/Download/app.apk"*
• *"Run this Python code on the device: print('Hello from Android!')"*

Links

• 🌐 Website: yydsauto.com
• 📦 npm: yyds-auto-mcp
• 🐙 GitHub: yyds-auto-mcp