mobile-mcp-optimization

name	mobile-mcp-optimization
description	Enforce the use of mcp_mobile_multi_action to reduce latency and improve reliability during Mobile MCP UI automation.

Mobile MCP Optimization: Multi-Action

When automating mobile UI flows using the mobile-mcp tools, individual tool calls for actions like tapping, typing, scrolling, and waiting introduce significant network and cognitive processing latency between each step.

To optimize execution speed and drastically reduce latency, you MUST use the mcp_mobile_multi_action tool instead of making sequential calls to mcp_mobile_snapshot_and_act or other single-action tools whenever possible.

When to Use multi_action

Group actions together when:

1. Entering data: Tap on an input field, wait for the keyboard to appear, type text, and press Enter/Done.
2. Navigating menus: Tap a menu button, wait for the menu to open, and tap a menu item.
3. Filling forms: Sequence taps on multiple fields, typing, and the final submit button all in one go.
4. Scrolling and interacting: Swipe to reveal an element, wait for the animation, and then tap it.
5. Dismissing alerts/modals: Tap OK/Cancel on an alert, then proceed with the flow.

How to use multi_action

Instead of multiple LLM round-trips:

1. LLM decides to tap "Search".
2. Wait for tool result...
3. LLM decides to type "hello world".
4. Wait for tool result...
5. LLM decides to press "ENTER".

Plan the entire macro sequence and execute it in one call:

{
  "device": "<device_id>",
  "actions": [
    { "action": "tap_text", "text": "Search", "delayAfter": 1000 },
    { "action": "type", "text": "hello world", "delayAfter": 500 },
    { "action": "press", "button": "ENTER", "delayAfter": 1500 }
  ]
}

Supported Actions within multi_action:

• tap (requires x, y)
• tap_text (requires text - Preferred over coordinates when possible)
• tap_id (requires id - Safest option if resource-ids are available)
• swipe (requires direction: up, down, left, right)
• type (requires text)
• press (requires button, e.g., HOME, BACK, ENTER, VOLUME_UP, VOLUME_DOWN)
• long_press (requires x, y, duration in ms)
• double_tap (requires x, y)
• drag (requires startX, startY, endX, endY, duration in ms)
• wait (requires ms - Explicit pause)
• wait_for_text (requires text, optional timeout in ms, default 5000)
• wait_for_id (requires id, optional timeout in ms, default 5000)

Best Practices for multi_action

1. Always use delayAfter for UI transitions: When an action causes an animation (like opening a keyboard, navigating a screen, or expanding a dropdown), the next action might fail if executed instantly. Add "delayAfter": 500 (or 1000 for slower transitions) to the action that triggers the animation.
2. Use wait_for_text / wait_for_id for network loads: If navigating to a new screen requires a network request, use a wait_for_* action to ensure the screen is fully loaded before continuing the sequence. This is more robust than a fixed delayAfter.
3. Chain related tasks: If you know you need to fill 3 text fields, do them all in one multi_action. Don't snapshot the screen between every single field if you already know their layout.
4. Use skipSnapshot: true for blind sequences: If the sequence is just pressing HOME or navigating back multiple times and you don't immediately need the resulting UI tree, set skipSnapshot: true on the mcp_mobile_multi_action call to save bandwidth and time.