AI-powered mobile device control chat interface. Connects to Android devices via MCP (Model Context Protocol) and uses LLM + VLM to understand and interact with screen content.
- Natural language control of Android devices (tap, swipe, type, etc.)
- AI vision (VLM) for screen understanding when UI nodes aren't enough
- Real-time SSE event streaming with web UI
- OpenAI-compatible API backend support
User Input → myt-chat (Go) → LLM API → Tool Calls → MCP Server → Android Device
↑ |
└── VLM (screenshot analysis) ←────┘
- Go 1.23+
- A running MCP server (e.g. myt-mcp)
- An OpenAI-compatible LLM API endpoint
- (Optional) A VLM API endpoint for screen recognition
go build -o myt-chat .
Create config.json in the same directory as the binary:
{
"ai": {
"api_url": "http://your-llm-api/v1/chat/completions",
"api_key": "your-api-key",
"api_model": "your-model-name"
},
"vlm": {
"api_url": "http://your-vlm-api/v1/chat/completions",
"api_key": "your-vlm-key",
"api_model": "your-vlm-model"
},
"mcp": {
"server_path": "./myt-mcp"
},
"server": {
"port": 8080
},
"log": {
"ai": true,
"vlm": true,
"mcp": true
}
}
# With config file
./myt-chat
# With flags
./myt-chat -api http://your-api/v1/chat/completions -key your-key -model your-model -mcp ./myt-mcp
# With env vars
export AI_API_URL=http://your-api/v1/chat/completions
export AI_API_KEY=your-key
./myt-chat
Open http://localhost:8080 in your browser, enter device IP, click connect, then chat.
| Endpoint | Method | Description |
|---|---|---|
/api/chat | POST | Send a chat message |
/api/abort | POST | Abort current operation |
/api/events | GET | SSE stream of chat events |
/api/tools | GET | List available MCP tools |
/api/connect | POST | Connect to a device |
MIT
35/F,Tencent Building,Kejizhongyi Avenue,Nanshan District,Shenzhen
