robotframework-maestrolibrary

Robot Framework keywords for mobile apps, driven by Maestro. If you know AppiumLibrary, you already know the keyword names, the arguments and the strategy=value locators. There's no Appium server and no capabilities to manage. Maestro waits for the UI on every action.

Contents

• Requirements
• Install
• Example
• Locators
• Coming from AppiumLibrary
• Android logs and mirroring
• iOS (WIP, needs testing)
• Converting between Maestro flows and Robot tests
• Studio: record a test on a live device
• Maestro CLI coverage
• Development

Requirements

• Python 3.12 or newer
• Maestro CLI on PATH (maestro --version) and Java 17+
• A running Android emulator or connected Android device, or an iOS simulator on a Mac (WIP, see iOS)
• Android: adb (platform-tools) on PATH. Open Application uses it to wait for the app's window; Execute Adb Shell, screen recording and the logcat files need it. Without it, Open Application warns and returns as soon as Maestro does.
• Optional, for screen mirroring and Studio's live view: scrcpy on PATH and a display.
• maestro, adb and scrcpy are taken only from PATH's absolute folders, never the working folder, and only if they answer like the real tool (adb version, scrcpy --version, and Maestro's MCP server naming itself maestro). Anything else is stopped and refused.

Install

pip install robotframework-maestrolibrary

For development: pip install -e .

Bugs and ideas: issues. Releases: a vX.Y.Z tag matching __version__ publishes to PyPI from GitHub Actions (docs/distribution.md). MIT: see LICENSE.

Example

*** Settings ***
Library    MaestroLibrary    timeout=10s

*** Test Cases ***
Search Settings
    Open Application    com.android.settings
    Click Text    Search settings    exact_match=True
    Input Text    id=com.google.android.settings.intelligence:id/open_search_view_edit_text    wifi
    Wait Until Page Contains    Wi-Fi
    [Teardown]    Close All Applications

Locators

text=, accessibility_id=, id= (literal, whole value), regex=, id_regex= (Maestro regular expressions), point=50%,50%. A bare value means text. Flutter apps get ids from Semantics(identifier: ...). xpath, class, android, ios, predicate, chain, css, name and identifier fail with a clear message.

Coming from AppiumLibrary

Same names and arguments: Close/Activate/Terminate Application, Close All Applications, Go Back, Go To Url, Get Source, Log Source, Execute Adb Shell, Click Element, Click Text, Input Text, Input Password, Input Text Into Current Element, Clear Text, Hide Keyboard, Page Should (Not) Contain Text/Element, Element Should Be Visible/Enabled/Disabled, Text Should Be Visible, Element Text Should Be, Element Should (Not) Contain Text, Get Text, Get Element Attribute, Scroll Element Into View, Expect Element, Expect Text, the five Wait Until keywords, Swipe, Swipe By Percent, Scroll Down/Up, Tap, Long Press, Press Keycode, Capture Page Screenshot, Register Keyword To Run On Failure, Set Location, Landscape, Portrait, Click Alert Button (iOS: WIP).

Differences:

• Open Application <app id> clear_state=False stop_app=True: no remote URL or capabilities. Like Appium, it returns once the app has window focus.
• Set/Get Appium Timeout become Set/Get Maestro Timeout.
• Press Keycode supports the keycodes that Maestro can send. Press Key takes Maestro key names.
• Should/Get keywords check the settled screen once. Wait/Expect keywords and every action wait.
• Screenshots are JPEG (maestro-screenshot-<n>.jpg, or EMBED).
• Swipe duration takes a time (300ms); a bare number is milliseconds, as in AppiumLibrary.
• speed=0.5s (import argument) pauses before every Maestro command, like SeleniumLibrary's speed. Use it when a slow or overloaded emulator stops responding.
• Start Screen Recording takes time_limit (AppiumLibrary: timeLimit), 3 minutes by default. Stop Screen Recording saves filename as given (AppiumLibrary appends .mp4). Both are Android only (adb screenrecord): Maestro ends its own recording with each flow, and every keyword is its own flow. Put Stop in a teardown.
• Set Location ignores altitude.

Maestro-only keywords:

• Run Flow: a YAML file, inline commands, or a directory of flows with include_tags / exclude_tags, plus env variables.
• App state: Clear Application State, Kill Application (the system killing a background app), Set Application Permissions (camera=deny, all=allow), Open Application permissions=....
• Device: Travel (move the location along points), Set Orientation, Set Airplane Mode, Set Dark Mode, Add Media (put images or videos in the gallery).
• Visual: Capture Element Screenshot (PNG), Screenshot Should Match (compare with a reference PNG, whole screen or one element, with a threshold).
• Wait For Animation To End, Is Keyboard Shown.
• Android: Start Screen Mirroring / Stop Screen Mirroring (scrcpy).

Not ported: webview contexts, xpath, multiple app aliases, touch id, WebElement keywords, Drag And Drop, Flick, sleep-between-wait-loop settings, and the iOS keywords Lock, Input Value and Tap With Number Of Taps.

Android logs and mirroring

Logcat: on Android, every test's adb logcat is streamed to logcat/<n>-<test>.txt in the output directory and linked in the log, so a crash's log survives the device's small ring buffers. Suite setup and teardown get a file named after the suite. Needs adb; turn it off with logcat=False.

Mirroring: Start Screen Mirroring shows the device screen in a scrcpy window on this computer, and Stop Screen Mirroring closes it (a no-op when nothing runs, so it fits a teardown). mirror=True opens it for the whole run instead. The window is view only; control=True lets the mouse and keyboard through. Extra arguments go to scrcpy as given, for example --max-size=1024 or --record=run.mp4 for a recording longer than Start Screen Recording's 3 minutes.

Import arguments, set once in a shared resource file:

*** Settings ***
Documentation     common.resource: the library and its settings, for every suite.
Library           MaestroLibrary    timeout=10s    device=${DEVICE}    logcat=${LOGCAT}    mirror=${MIRROR}

*** Variables ***
${DEVICE}         ${None}     # the first connected device
${LOGCAT}         ${True}     # logcat/<n>-<test>.txt per test
${MIRROR}         ${False}    # a scrcpy window for the whole run

A suite that uses it, with mirroring and a recording for one test:

*** Settings ***
Resource          common.resource
Suite Setup       Open Application    com.example.app
Suite Teardown    Close All Applications

*** Test Cases ***
Login Works
    Input Text    id=username    demo
    Input Password    id=password    ${PASSWORD}
    Click Element    text=Log in
    Wait Until Page Contains    Welcome

Checkout While Watching
    [Setup]    Start Screen Mirroring    --max-size=1024    --always-on-top
    Start Screen Recording    time_limit=60s
    Click Element    text=Checkout
    Wait Until Page Contains    Order placed
    [Teardown]    Run Keywords    Stop Screen Recording    checkout.mp4
    ...           AND    Stop Screen Mirroring

robot -d results tests/                                      # logcat on, no window
robot -d results --variable MIRROR:True tests/               # watch the whole run
robot -d results --variable DEVICE:emulator-5554 --variable LOGCAT:False tests/

After a crash, open the test in log.html and follow its logcat link, or read results/logcat/<n>-<suite>.<test>.txt.

iOS (WIP, needs testing)

Nothing here has run on iOS yet: there was no Mac to test on. Maestro 2.11.0 drives iOS simulators only (macOS with Xcode); it fails physical iPhones with "Physical iOS devices are not yet supported". Using an iOS device logs one warning that iOS support is WIP.

Android only, failing at once on iOS (from Maestro's docs): Go Back (iOS has no back button), Press Key Back / Power / Tab and Press Keycode 4, Set Airplane Mode (simulators have none), Execute Adb Shell, Is Keyboard Shown.

WIP, written from Maestro's and Apple's docs, needs testing on a simulator:

• the per-test log files stream the simulator's log (xcrun simctl spawn <udid> log stream);
• Start/Stop Screen Recording uses xcrun simctl io <udid> recordVideo;
• Start Screen Mirroring does nothing (the Simulator window shows the screen);
• Clear Keychain, Click Alert Button, Hide Keyboard key_name=Done;
• xcrun is checked like adb (xcrun --version must answer xcrun version <n>).

Not checked on iOS yet: Kill Application, permission names for Set Application Permissions, which attributes Get Element Attribute sees, and how id= matches.

Converting between Maestro flows and Robot tests

Record a flow in Maestro Studio, save its YAML, and convert it to a Robot test:

pip install "robotframework-maestrolibrary[convert]"
python -m MaestroLibrary.flow2robot settings_search.yaml > settings_search.robot

A Studio flow such as

appId: com.android.settings
---
- launchApp
- assertVisible: "Search Settings"
- tapOn:
    text: "Apps"
    index: 0
- back
- tapOn:
    id: "com.android.settings:id/search_action_bar"
- inputText: "wifi"
- takeScreenshot: search
- hideKeyboard
- back
- stopApp

becomes

*** Settings ***
Library    MaestroLibrary

*** Test Cases ***
Settings Search
    Open Application    com.android.settings
    Wait Until Page Contains Element    text\=Search Settings
    Run Flow    - {tapOn: {text: Apps, index: 0}}
    Go Back
    Input Text    id_regex\=com.android.settings:id/search_action_bar    wifi
    Capture Page Screenshot    search.jpg
    Hide Keyboard
    Go Back
    Terminate Application    com.android.settings

(\= keeps Robot from reading id=... as a named argument.)

Commands with a keyword become that keyword (tapOn then inputText becomes one Input Text); the rest, such as a selector with index, become an inline Run Flow line. Maestro matches text as a regular expression, so a value with regex characters (an id with dots, too) comes out as regex= or id_regex=, which keeps its meaning. Move the locators into variables before the test joins a suite.

Robot tests to Maestro flows

A Robot test is only a flow once it runs: variables, loops and user keywords resolve at run time. So the library records instead of converting. With record_flows=True, every Maestro command a test sends goes to flows/<n>-<test>.yaml in the output directory, linked in the log:

robot -v RECORD:True -d results tests/      # Library  MaestroLibrary  record_flows=${RECORD}
maestro test -e PASSWORD=... results/flows/1-Log_In.yaml

Loops come out unrolled and Run Flow lines are kept; Input Password is recorded as ${PASSWORD}. Snapshot checks (Page Should Contain Element, Get Text, ...), screenshots, adb and Python steps are not Maestro commands, so they are not in the flow; waiting checks such as Wait Until Page Contains Element are (extendedWaitUntil).

Studio: record a test on a live device

pip install "robotframework-maestrolibrary[studio]"      # PySide6, PyAV and grpcio
python -m MaestroLibrary.studio --app com.android.vending

Or, without Python, download the Windows .exe, Linux .tar.gz or macOS .dmg from the GitHub release (unsigned for now; the release text says how to open them). Maestro, adb and scrcpy are still needed on the machine.

A window shows the device screen live, and every action you take on it runs on the device while the matching Robot line is recorded:

Inspect mode selects instead of acting: the element's suggested locators (with how many elements each matches), its attributes and the source tree. Undo, Clear, Copy and Save work on the recorded lines; Record pauses recording while actions still run. Move the locators into variables and file the GAP lines as test id requests before the test joins a suite.

• The live view needs scrcpy on PATH and an Android device; it streams the device's hardware H.264 encoder at up to 1080 px (--max-size; --video-encoder picks another encoder from scrcpy --list-encoders). Without it, or on an iOS simulator, the view refreshes with a screenshot after each step.
• On Android with adb, taps, long presses, swipes, Back and plain typing (letters, digits, spaces and . , _ @ : + = / -) go to the device with adb shell input and are recorded at once. Other text and secrets go through Maestro, as do assertions and Launch, which waits for the screen to settle. The recorded line is the same either way, and the locator is checked when the test runs. Measured on an Android 15 phone: the line is recorded 0.13 to 0.17 s after a click and the screen changes 0.33 to 0.43 s after it (a Maestro tap took 4 to 7 s); typing takes about 0.3 s against 4 s through Maestro.
• On Android, the element tree is read from Maestro's on-device driver (its gRPC viewHierarchy, through an adb forward): 0.45 s on an Android 15 phone against 2.2 s for MCP's inspect_screen, which stays the fallback.
• With several devices connected, the device box in the toolbar lists them (phones and emulators marked apart); picking one moves Studio to it in place and keeps the recorded lines.
• The theme button in the toolbar cycles System, Light and Dark; the choice is remembered between runs.
• Shortcuts: Ctrl+1 Act, Ctrl+2 Inspect, Ctrl+Z undo, Ctrl+S save, Delete removes the selected recorded line.
• Keep Robot runs off the device while Studio is open: Maestro allows one session per device.
• Linux needs Qt's system libraries (Debian/Ubuntu: libxcb-cursor0); Windows and macOS need nothing extra.
• Device text is shown as plain text, and recorded arguments escape Robot variable syntax, so an app cannot inject code into the saved test.

Maestro CLI coverage

The library talks to maestro mcp. What the other maestro subcommands do, and where it lives here:

Flow commands without a keyword, still reachable with Run Flow: the AI assertions (need a Maestro Cloud key), inputRandom*, clipboard commands (the copied text stays inside one flow), repeat/retry (use Robot's FOR and Wait Until Keyword Succeeds), runScript/evalScript, toggleAirplaneMode, assertDarkMode/assertLightMode, clearKeychain (iOS), doubleTapOn.

Each maestro mcp process reinstalls Maestro's driver app on the device once, when it first connects (Maestro hard-codes it). The library keeps one process for the whole run.

Tip: to let a coding agent look at the app while you write tests, register the same server with it, e.g. claude mcp add maestro -- maestro mcp for Claude Code. It opens its own session on the device, so keep it idle while a Robot run is going.

Development

python -m unittest discover -s utest                        # no device needed
robot --pythonpath src -d results atest                     # needs a running Android emulator
python -m robot.libdoc --pythonpath src MaestroLibrary results/MaestroLibrary.html