android-device-automation

Android Device Automation

CRITICAL RULES — VIOLATIONS WILL BREAK THE WORKFLOW:

1. Never run midscene commands in the background. Each command must run synchronously so you can read its output (especially screenshots) before deciding the next action. Background execution breaks the screenshot-analyze-act loop.
2. Run only one midscene command at a time. Wait for the previous command to finish, read the screenshot, then decide the next action. Never chain multiple commands together.
3. Allow enough time for each command to complete. Midscene commands involve AI inference and screen interaction, which can take longer than typical shell commands. A typical command needs about 1 minute; complex act commands may need even longer.
4. Always report task results before finishing. After completing the automation task, you MUST proactively summarize the results to the user — including key data found, actions completed, screenshots taken, and any relevant findings. Never silently end after the last automation step; the user expects a complete response in a single interaction.

Automate Android devices using npx -y @midscene/android@1. Each CLI command maps directly to an MCP tool — you (the AI agent) act as the brain, deciding which actions to take based on screenshots.

What act Can Do

Inside a single act call on Android, Midscene can tap, double-tap, long-press, type, clear text, scroll or swipe in any direction, pull to refresh, drag items, zoom with two fingers, press keys, and use system navigation such as Back, Home, or recent apps while working from the current visible screen.

Prerequisites

Midscene requires models with strong visual grounding capabilities. The following environment variables must be configured — either as system environment variables or in a .env file in the current working directory (Midscene loads .env automatically):

MIDSCENE_MODEL_API_KEY="your-api-key"
MIDSCENE_MODEL_NAME="model-name"
MIDSCENE_MODEL_BASE_URL="https://..."
MIDSCENE_MODEL_FAMILY="family-identifier"

Example: Gemini (Gemini-3-Flash)

Example: Qwen 3.5

Example: Doubao Seed 2.0 Lite

Commonly used models: Doubao Seed 2.0 Lite, Qwen 3.5, Zhipu GLM-4.6V, Gemini-3-Pro, Gemini-3-Flash.

If the model is not configured, ask the user to set it up. See Model Configuration for supported providers.

Commands

Common Android CLI Flags

Use these flags on commands that create or use the Android agent, such as connect, take_screenshot, act, assert, and tap:

• --device-id <id>: Target a specific Android device from adb devices.
• --use-scrcpy: Enable scrcpy accelerated screenshots. Use this when normal screenshot capture is slow or unstable. Because CLI invocations are stateless, pass this flag on each Android Midscene command where you want scrcpy-based screenshots.

Connect to Device

Launch an App or URL

Use the dedicated launch step when you want a deterministic starting point before the rest of the task:

Run a Raw Android Shell Command

Use this when the task needs lower-level device control that is not best expressed as a visible UI interaction:

This is forwarded to adb shell on the connected device. In practice, the underlying command is adb -s <deviceId> shell dumpsys battery and some environments may also include the default ADB server port, such as adb -P 5037 -s <deviceId> shell dumpsys battery.

Take Screenshot

After taking a screenshot, read the saved image file to understand the current screen state before deciding the next action.

Perform Action

Use act to interact with the device and get the result. It autonomously handles all UI interactions internally — tapping, typing, scrolling, swiping, waiting, and navigating — so you should give it complex, high-level tasks as a whole rather than breaking them into small steps. Describe what you want to do and the desired effect in natural language:

Assert Current Screen State

Use assert to verify that the current screen satisfies a natural language condition. It does not perform UI actions; it checks the visible screen state and passes only when the assertion is true. Use this for validation, QA checks, and final state verification after act.

By default a failed assertion throws an AI-generated reason. Pass --message to throw a custom error message instead, which is useful for surfacing the intended outcome in QA and CI logs.

When the assertion needs to compare against a reference image (icon, logo, screenshot), pass --image for the URL/path and --image-name for its display name. Each --image may be an http(s) link, a data: URI, or a local file path. Repeat both flags in matching order when you need to attach more than one image. Add --convertHttpImage2Base64 true when the model cannot reach the URL directly. Requires @midscene/android@1.9.0+.

Record and Assert Transient UI

Use a recording when the state to verify may disappear before a current-screen assertion runs, such as a toast, loading banner, animation, or transition:

Pass target flags, capture flags, and --output to record start, then wait for Recording. Press Ctrl+C to stop and save. Keep the recorder as a foreground process in its dedicated terminal; never add shell &. Perform the interaction manually or from a second terminal, send Ctrl+C to the recorder, and wait until it prints the saved path before asserting. --use-scrcpy provides the faster continuous frame source; without it, recording falls back to periodic screenshots. Optional capture flags are --interval-ms, --max-frames, and --watchdog-ms; --max-frames caps sampled frames, and the manifest may contain one additional final representative frame. The default watchdog finalizes and saves the recording after five minutes, while --watchdog-ms 0 disables that safety limit. The output is a JSON manifest plus an adjacent <name>.frames image directory, not an encoded video or archive. The manifest contains relative JPEG/PNG paths and no base64 image bodies. Keep or move the JSON file and image directory together, and pass the JSON path to assert --record. Use ordinary assert without --record when only the current screen matters.

Use a Reference Image for Precise Targeting

When the user provides a screenshot, icon, logo, or reference image and wants an exact visual match, prefer tap --locate instead of a generic act --prompt. Pass --locate as JSON. The prompt describes the target, images supplies named reference images, and convertHttpImage2Base64: true is useful when the image URL may not be directly accessible to the model.

The same locate JSON shape also works for other commands that accept a locate parameter.

Disconnect

Consume Report Files

The generated HTML report is recommended for human reading first. It includes step-by-step execution details and replay videos for each operation, which makes it much easier to understand what happened and troubleshoot problems.

If another skill or tool needs to consume the report, first convert it with report-tool from the same platform CLI package. Prefer Markdown for LLM-based workflows. Use JSON when the report needs to be processed programmatically.

Workflow Pattern

Since CLI commands are stateless between invocations, follow this pattern:

1. Connect to establish a session
2. Launch the target app and take screenshot to see the current state, make sure the app is launched and visible on the screen.
3. Execute action using act to perform the desired action or target-driven instructions. Use assert for the resulting screen state, or keep record start --output ... running in a dedicated terminal during transient-state workflows, stop it with Ctrl+C, and then use assert --record.
4. Disconnect when done
5. Report results — summarize what was accomplished, present key findings and data extracted during the task, and list any generated files (screenshots, logs, etc.) with their paths

Best Practices

1. Bring the target app to the foreground before using this skill: For best efficiency, launch the app using ADB (e.g., adb shell am start -n <package/activity>) before invoking any midscene commands. Then take a screenshot to confirm the app is actually in the foreground. Only after visual confirmation should you proceed with UI automation using this skill. ADB commands are significantly faster than using midscene to navigate to and open apps.
2. Be specific about UI elements: Instead of vague descriptions, provide clear, specific details. Say "the Wi-Fi toggle switch on the right side" instead of "the toggle".
3. Describe locations when possible: Help target elements by describing their position (e.g., "the search icon at the top right", "the third item in the list").
4. Never run in background: Every midscene command must run synchronously — background execution breaks the screenshot-analyze-act loop.
5. Batch related operations into a single act command: When performing consecutive operations within the same app, combine them into one act prompt instead of splitting them into separate commands. For example, "open Settings, tap Wi-Fi, and toggle it on" should be a single act call, not three. This reduces round-trips, avoids unnecessary screenshot-analyze cycles, and is significantly faster.
6. Use scrcpy when screenshot capture needs acceleration: Add --use-scrcpy to each relevant command when normal Android screenshots are slow, flaky, or blocked by the environment.
7. Choose the right verification window: Use assert --prompt "..." for the current screen. For a toast, banner, animation, or transition, run record start --output ... in a dedicated terminal, perform the interaction, stop recording with Ctrl+C, wait for the saved-path message, then pass the artifact to assert --record.
8. Always report results after completion: After finishing the automation task, you MUST proactively present the results to the user without waiting for them to ask. This includes: (1) the answer to the user's original question or the outcome of the requested task, (2) key data extracted or observed during execution, (3) screenshots and other generated files with their paths, (4) a brief summary of steps taken. Do NOT silently finish after the last automation command — the user expects complete results in a single interaction.
9. Prefer tap --locate when a reference image is provided: If the user shares a screenshot, icon, or logo and wants that exact visual target, use tap --locate with a multimodal locate JSON object such as { "prompt": "...", "images": [...] } instead of relying only on act --prompt.

Example — Popup menu interaction:

Example — Form interaction:

Improve Precision (Deep Locate / Deep Think)

Two optional global flags help when Midscene struggles with a task. Put them anywhere in the command (before or after the sub-command); once set, the relevant operations use them by default, so you don't pass a per-call parameter.

• --deep-locate — spends an extra round of visual reasoning to pinpoint the target element. Use it when an action interacts with the wrong spot (location drift / offset). It applies to every operation that locates an element, including tap --locate and the locating that happens inside act.
• --deep-think — plans act with deeper reasoning (richer context and sub-goal decomposition). Use it for complex, multi-step act instructions; it only affects planning.

Both trade a little speed for better results, and you can combine them.

Troubleshooting