# Gaze

> Open-source Android device manager with built-in MCP (Model Context Protocol) server for AI-powered automation.

Gaze is a cross-platform desktop application (macOS, Windows, Linux) built with Go + React on the Wails framework. It provides comprehensive Android device management through ADB, featuring a built-in MCP server that exposes 54 tools for AI assistants like Claude Desktop, Claude Code, and Cursor.

## MCP Server

Gaze includes a built-in MCP server that allows AI clients to control Android devices programmatically.

- Transport: stdio (launch with `--mcp` flag)
- Tools: 54 tools across 8 categories
- Resources: 5 resource endpoints
- Compatible with: Claude Desktop, Claude Code (CLI), Cursor, and any MCP-compatible client

### Setup

Claude Desktop: Add to `claude_desktop_config.json`:
```json
{ "mcpServers": { "gaze": { "command": "/Applications/Gaze.app/Contents/MacOS/Gaze", "args": ["--mcp"] } } }
```

Claude Code: Run `claude mcp add gaze -- /Applications/Gaze.app/Contents/MacOS/Gaze --mcp`

Cursor: Add to `.cursor/mcp.json`:
```json
{ "mcpServers": { "gaze": { "command": "/Applications/Gaze.app/Contents/MacOS/Gaze", "args": ["--mcp"] } } }
```

## MCP Tool Categories

- Device Management (11 tools): device_list, device_info, device_connect, device_disconnect, device_pair, device_wireless, device_ip, adb_execute, aapt_execute, ffmpeg_execute, ffprobe_execute
- App Management (8 tools): app_list, app_info, app_start, app_stop, app_running, app_install, app_uninstall, app_clear_data
- Screen Control (4 tools): screen_screenshot, screen_record_start, screen_record_stop, screen_recording_status
- UI Automation (6 tools): ui_hierarchy, ui_search, ui_tap, ui_swipe, ui_input, ui_resolution
- Session Management (6 tools): session_create, session_end, session_active, session_list, session_events, session_stats
- Workflow Automation (12 tools): workflow_list, workflow_get, workflow_create, workflow_update, workflow_delete, workflow_run, workflow_stop, workflow_pause, workflow_resume, workflow_step_next, workflow_status, workflow_execute_step
- Network Proxy (3 tools): proxy_start, proxy_stop, proxy_status
- Video Processing (4 tools): video_frame, video_metadata, session_video_frame, session_video_info

## MCP Resources

- gaze://devices — List of connected Android devices (JSON)
- gaze://devices/{deviceId} — Detailed device information
- gaze://sessions — Active and recent tracking sessions (JSON)
- workflow://list — All saved workflows (JSON)
- workflow://{workflowId} — Workflow details with steps

## Core Features

- Screen mirroring and recording via scrcpy with audio forwarding
- File manager with drag-and-drop upload/download
- App management: install, uninstall, inspect APK details
- Real-time logcat viewer with package/tag/level filters and regex search
- HTTP/HTTPS network proxy with MITM decryption and auto certificate install
- UI hierarchy inspector with element picker
- Visual workflow automation engine with branching and variables
- Session-based unified event tracking timeline (logcat, network, device state, app lifecycle)
- Multi-device support with wireless ADB

## What AI Can Do with Gaze

Using MCP tools, AI assistants can:
- Automate Android UI testing by tapping, swiping, and inputting text
- Take screenshots and analyze UI hierarchy to understand app state
- Launch/stop apps and manage app lifecycle
- Capture and inspect network traffic via proxy
- Create and execute automated workflow scripts
- Monitor device sessions with logcat, performance, and state events
- Process video recordings for analysis

## Links

- Website: https://gaze.nicetooo.com
- GitHub: https://github.com/nicetooo/Gaze
- Download: https://github.com/nicetooo/Gaze/releases/latest
- Full tool reference: https://gaze.nicetooo.com/llms-full.txt