version: "1.0.1" name: mobai description: Control Android and iOS mobile devices through the MobAI HTTP API. Use this skill when the user asks to interact with mobile devices, take screenshots, tap elements, type text, swipe, launch apps, or automate mobile tasks. allowed-tools: mcp__mobai-http__http_request

MobAI Usage Rules (IMPORTANT)

When automating mobile devices, ALWAYS use this order:

1. mobai:native-runner - For ANY native app UI automation (tapping, typing, swiping, observing UI)
2. mobai:web-runner - For ANY web browser or WebView automation
3. mobai:agent - For complex multi-step tasks requiring AI reasoning
4. Raw HTTP API - ONLY for: listing devices, starting/stopping bridge

ALWAYS try DSL subagents first. Raw HTTP API for tap/type/swipe/screenshot/ui-tree is a LAST RESORT.

Screenshots: When using the API, screenshots are automatically saved to /tmp/mobai/screenshots/ and the path is returned. Use the Read tool to view them.

MobAI Device Control

This skill enables you to control Android and iOS devices through the MobAI HTTP API running locally.

Sub-Agent Architecture

For complex automation tasks, use a hierarchical approach with specialized sub-agents:

When to Use Sub-Agents

Native Runner (/native-runner)

Use for native mobile apps - apps that use platform UI components:

• Settings app, Mail, Photos, Calendar
• Third-party apps (Instagram, WhatsApp, Uber)
• Any app where you need to tap UI elements by accessibility predicates

Uses DSL batch execution with element predicates for robust automation.

How to spawn:

Use the native-runner skill to accomplish: [subgoal description]
Device ID: [deviceId]

Web Runner (/web-runner)

Use web-runner when you need to interact with DOM content inside a web page or WebView:

• Native-runner failed with NO_MATCH on web page elements
• You need CSS selectors to target specific HTML elements
• You need to execute JavaScript on the page
• You need to read/manipulate DOM content programmatically

IMPORTANT: iOS Simulators NOT supported - Web context requires a physical iOS device. Use native-runner for simulators.

Try native-runner first - it works for most web page interactions via accessibility tree.

DO NOT use for browser UI elements (address bar, tabs, back button) - those are native!

Uses DSL batch execution with CSS selectors and JavaScript.

How to spawn:

Use the web-runner skill to accomplish: [subgoal description]
Device ID: [deviceId]

Example: Complex Task Decomposition

User request: "Log into Twitter, search for 'AI news', and screenshot the results"

Step 1: List devices to get device ID (direct API call) Step 2: Launch Twitter app (direct API call) Step 3: Spawn native-runner: "Tap the search tab and enter 'AI news'" Step 4: Wait for results (determine if it's native or web) Step 5: If web results: spawn web-runner: "Scroll to see results" Step 6: Take screenshot (direct API call)

How to Make API Calls

Use the mcp__mobai-http__http_request tool to make HTTP requests:

{
  "method": "GET",
  "url": "http://127.0.0.1:8686/api/v1/devices"
}

For POST/PUT/PATCH requests with a body:

{
  "method": "POST",
  "url": "http://127.0.0.1:8686/api/v1/devices/{id}/dsl/execute",
  "body": "{\"version\":\"0.2\",\"steps\":[{\"action\":\"observe\",\"context\":\"native\"}]}"
}

Parameters:

• method: HTTP method (GET, POST, PUT, PATCH, DELETE)
• url: Full URL to request
• body: Request body as JSON string (for POST/PUT/PATCH)
• headers: Optional additional headers
• timeout: Request timeout in milliseconds (default: 600000 = 10 minutes)

API Base URL

http://127.0.0.1:8686/api/v1

No authentication is required. The API runs on localhost only.

Response Format

Success responses:

{"success": true, "data": {...}}

Error responses:

{"error": "error message", "code": "optional_code", "details": "optional_details"}

Quick Reference - Common Operations

Device Management (Direct API)

GET /devices                    # List all connected devices
GET /devices/{id}               # Get specific device info
GET /devices/{id}/screenshot    # Capture screenshot (base64 PNG)
GET /devices/{id}/apps          # List installed apps (or use DSL observe with include: ["installed_apps"])

Bridge Control (Direct API)

POST /devices/{id}/bridge/start # Start on-device bridge (60s timeout)
POST /devices/{id}/bridge/stop  # Stop on-device bridge

DSL Batch Execution (Preferred for Automation)

POST /devices/{id}/dsl/execute  # Execute DSL script with retries

Example DSL script:

{
  "version": "0.2",
  "steps": [
    {"action": "observe", "context": "native", "include": ["ui_tree"]},
    {"action": "tap", "predicate": {"text_contains": "Settings"}},
    {"action": "observe", "context": "native", "include": ["ui_tree"]}
  ],
  "on_fail": {"strategy": "retry", "max_retries": 2}
}

Scrollable List Operations (Direct API)

POST /devices/{id}/scroll-until-visible  # Scroll to find element
POST /devices/{id}/collect-list          # Collect all list items

AI Agent (Direct API)

POST /devices/{id}/agent/run    # {"task": "...", "agentType": "toolagent"}

App Management (Direct API)

POST /devices/{id}/kill-app          # Force-kill app: {"bundleId": "..."}
DELETE /devices/{id}/apps/{bundleId} # Uninstall app by bundle ID

Location Simulation (Direct API)

POST /devices/{id}/location          # Set GPS: {"lat": 40.71, "lon": -74.00}
DELETE /devices/{id}/location        # Reset to real GPS

Performance Metrics (Via DSL)

Use metrics_start and metrics_stop DSL actions for performance testing:

{
  "version": "0.2",
  "steps": [
    {"action": "metrics_start", "types": ["system_cpu", "system_memory", "fps", "network", "battery"], "capture_logs": true, "label": "test"},
    {"action": "open_app", "bundle_id": "com.example.app"},
    {"action": "delay", "duration_ms": 5000},
    {"action": "metrics_stop", "format": "summary"}
  ]
}

Returns health score, anomalies, and recommendations. Metric types: system_cpu, system_memory, fps, network, battery, process.

Choosing Between Native and Web Mode

CRITICAL: Browser apps (Safari, Chrome) have TWO zones:

1. Browser chrome (NATIVE): address bar, tab bar, back/forward buttons, share button, bookmarks → use native-runner
2. Web content viewport (WEB): the actual webpage content inside the browser → use web-runner

Use Native Mode (native-runner) when:

• Working with native app UI (buttons, switches, lists)
• Interacting with browser chrome: URL bar, navigation buttons, tab switching
• Tapping by element predicate (text, type, label)
• UI tree shows native components (Button, TextField, Switch)

Use Web Mode (web-runner) ONLY when:

• You need to interact with the DOM content of a web page or WebView
• Filling HTML forms, clicking HTML links/buttons, or reading text from web pages
• You need CSS selectors to target elements rendered by HTML/CSS/JS
• Native taps aren't working because the content is rendered by WebKit/Blink engine

NEVER use web-runner for: browser chrome UI (address bar, tabs, navigation buttons) - always use native-runner for those!

Detection Tips:

1. Get UI tree first - if it shows web-like elements (WebView), the CONTENT is web mode
2. Browser UI elements (address bar, tabs) are ALWAYS native even when Safari/Chrome is open
3. If native tap fails on expected element inside webpage, try web mode

Important Notes

• Use DSL for automation - Batch execution is more efficient than individual API calls
• Direct API for queries - List devices, screenshots, app lists don't need DSL
• Always ensure the bridge is running (bridgeRunning: true) before automation
• Sub-agents use DSL internally - they handle batch execution for you
• See api-reference.md for full endpoint documentation