Unattended remote access and remote control

[Rambomst]

Turns the Android agent into a full unattended remote-control agent: background screen sharing and remote input, running as a persistent foreground service.

Unattended operation

• specialUse foreground service + boot receiver, so it runs and reconnects without the app open.
• Shared AgentController runtime for lifecycle; unattended setup prompts and settings.
• Consent follows the server's consent flags and the app's Automatic Consent setting, for desktop and files sessions alike. Prompts show in-app or as a notification with Approve/Deny, expire after the server's consent timeout (30 s default), and file requests are held until approved.

Remote control (previously view-only)

• Accessibility service provides background capture (takeScreenshot) and remote input.
• Pointer input is streamed as live gestures: click = tap, hold = long press, drag = live drag (scrolling, selection handles, moving icons), wheel = swipe, right-click = Back, middle-click = Home.
• Keyboard input into the focused field, password fields included, with Enter as the field's keyboard action.
• Android global actions from the web panel (Add on-screen input panel for Android devices in remote desktop views MeshCentral#8123): Back, Home, Recents, app drawer (Android 14+ action with a Home + swipe fallback for older launchers), notifications, quick settings, lock, power.
• Lock screen: the display is woken for a session, and a PIN or password can be typed blind since Android blanks the PIN pad from capture. Pattern locks are not supported.
• Clipboard: upload from the browser works in the background; download only works while the agent app is in the foreground (Android 10+ restriction).

Files and software

• Uploads work under scoped storage: MediaStore for the media folders and for Android 10, raw paths where Android allows them, clear errors on the Files tab otherwise, plus uploadhash/append/cancel support.
• Media folders are listed from the MediaStore index (several times faster on large camera rolls), downloads fall back to MediaStore rows, and the web file editor's block download protocol is implemented.
• Optional All files access (MANAGE_EXTERNAL_STORAGE) behind -PmeshAllFilesAccess=true, off by default because Google Play requires a use-case declaration for it.
• Software tab lists launchable apps without QUERY_ALL_PACKAGES.

Platform & tooling

• Merged latest upstream (Android SDK 37), file-download fix, edge-to-edge/system-bar handling.
• GitHub Actions PR build workflow; developer docs under docs/; German translations complete, lint clean.

Known limitations

• Sideloads are flagged by Play Protect (accessibility/remote-control category); install via adb install or enterprise deployment. On Android 13+ the accessibility toggle is a Restricted Setting unless installed via ADB/EMM.
• Unattended capture is ~2-3 fps (Android throttles AccessibilityService.takeScreenshot).
• Secure surfaces and the lock screen PIN pad are blanked by Android; multi-touch gestures such as pinch are not available.
• Without All files access, other apps' non-media files on Android 13+ can be listed but not transferred.

Disclaimer: developed with an AI coding assistant.

[Rambomst]

Ive had this running on two devices without any issues for the past week or so. My initial commit introduce some battery issues but that was resolved and now its back to basically using nothing.

[Rambomst]

A few fixes to the unattended consent flow, with Play Store approval in mind.

• Consent is properly enforced now. With Automatic Consent off, every session has to be approved (in-app dialog, or Approve/Deny on the notification), so there's no silent screen capture, which is the main thing Google flags for apps like this.
• Removed the battery optimization popup that auto-fired on launch, since auto-firing that intent is a known rejection trigger. It's only reachable from Settings now.
• Capture only runs during an active session and shows a persistent notification plus a per-session one, so it stays transparent rather than looking like spyware.

To hopefully get this through review we need to do this in Play Console:

• Complete the Accessibility API declaration and add a prominent in-app disclosure, framed as person-driven remote support as Google restricts accessibility use for automation as of the Jan 2026 policy change.
• Fill in the foreground service types declaration for the specialUse service which needs a short justification and a demo video. Something like this:

MeshCentral Agent is a remote management/support tool. The foreground service keeps a persistent connection to the management server open so an administrator can reach the device on demand for remote desktop, file transfer, and commands. It has to run continuously and indefinitely; it isn't a finite data sync, media playback, location job, or any other predefined types, so specialUse applies.

• There might also be a need to justify the battery optimization exemption, but that should be fine since this agent wouldn't work properly without it. If they did have an issue with it we can change it pretty easily but no harm in trying if this was already going through review.

How to sideload this with the accessibility changes:

Play Store > profile icon > Play Protect > gear/settings icon > toggle off "Scan apps with Play Protect" > install the APK > turn it back on after.

Then it blocks the accessibility access until we do this:
Settings > Apps > Mesh Central > App info⋮(top-right) > Allow restricted settings (authenticate), then the toggle works.

[GlitchedCod]

I hope that this get official one day, it's really a useful feature. The app itself without that is basically useless, it's not ok to ask a user how to go to the home menu, find all the apps, start the agent and then allow the screen sharing. Plus you can't see when a device is online on the MeshCentral dashboard since the app is not running in the background if it wasn't started at the launch of the device.

It would also be cool to add the "device administrator" rights to the app so that it can be set as the device owner, disable FRP on the device and so that the user can't uninstall it unless the device is wiped.

[Rambomst]

@GlitchedCod Yeah I hope this gets merged as well, I brought it up at the community meeting last night so hopefully it will get some traction.

As for your request around device administrator functionality, that would be out of scope for this PR. One thing of note around that, I don't think that sort of functionality would ever make it to the play store build, so if it was to be development it would be something that needs to get manually deployed to each device you want to manage via ADB.

[marclaporte]

I know @kambereBr is testing this PR. yay!

I look forward to this as I have an Android device which I need to manage from time to time :-)

[GlitchedCod]

I know @kambereBr is testing this PR. yay!

I look forward to this as I have an Android device which I need to manage from time to time :-)

That is so cool ! This is definitely gonna be a game changer. Fun fact I actually read this merge request yesterday to see if they were any update onto it.

[Rambomst]

I have created a MeshCentral PR which implements some UI changes to facilitate interactions with this agent, things like buttons to go home/back/recents menu/etc.

MeshCentral #8123

[kambereBr]

Hi @Rambomst

Tested this PR with the on-screen input panel PR (Ylianst/MeshCentral#8123) on a Pixel 3a, Android 12 (Snapdragon 670), installed via ADB from Android Studio.

Works well:

• App swiped away: stays online. Reboot: auto-starts and reconnects on its own.
• First-run setup handled the accessibility flow, and the settings status updated once granted.
• Background capture works with the app/AndroidAgent closed.
• Back, Home, Recent, Notifications, Quick, Lock and Power buttons work fine.

Issues:

• The "Waiting for user to grant access..." message stays on screen even after access is granted and I'm already controlling the device.
• The "Apps" button does nothing. The other actions work, so it seems specific to this one.
• Text selection has a small issue: the selection only expands if I release the mouse button inside the phone's screen. If the cursor is already outside the screen when I release it, nothing happens. Nothing unusual showed in logcat.
• File upload does not work (download works fine). Nothing unusual showed in logcat here either.

I recorded a short screencast showing the text selection and file upload issues, attached below.

Maybe out of scope of this PR, please let me know if that is the case:

• Long-press, right-click and drag-and-drop: couldn't find how to do these from the panel.
• Clipboard both ways (copy on web, paste on phone and vice versa) is not available.
• Opening/connecting to Files doesn't ask for consent on the device even when I choose "Ask Consent" on the web UI, I can access them silently. Should asking for consent be mandatory here?

I'll also test on an Android 16 device and send feedback if anything comes up.

Thank you for this PR, remote control is working really well.

MeshCentralAndroidAgent.mp4

[GlitchedCod]

Woah ! So cool already !

Good luck for the little patches needed.

[Rambomst]

@kambereBr Thanks again for the detailed test, all four should now be fixed.

• Banner: the agent never cleared "Waiting for user to grant access" once capture started. Both capture paths clear it now.
• Apps: the accessibility all-apps action is only wired to the launcher from Android 14, so on 12 and 13 it did nothing. The agent uses it on 14+ and checks the launcher came up, otherwise it goes Home and swipes up.
• Selection drag: input was replayed as one gesture on button release, and the browser never delivers a release outside the canvas. Input is now streamed live: drags move as you go, holding the button is a long press, right-click is Back, middle-click is Home. The viewer side in Add on-screen input panel for Android devices in remote desktop views MeshCentral#8123 now releases the button at the canvas edge.
• Upload: your target was Sdcard/Photos, a non-standard folder scoped storage refuses, and the agent said nothing. Uploads fall back to MediaStore where Android allows it, the Files tab now shows why a folder is refused, and re-uploading a same-named file no longer hangs.

Also in the same push: files sessions ask for consent per the server's flags and the app setting, clipboard upload to the device (download only works while the app is in front, an Android restriction), a Software tab inventory, and blind PIN unlock, since Android blanks the PIN pad from capture.

[Rambomst]

Update since the above comment:

• Keyboard: on Android 13+ typing now goes through the system input method, so fast typing stays in order and Ctrl+A/C/X/V, Shift+arrows, Delete and Tab work. Older releases keep the field-rewrite path with a guard against dropped characters.
• Dragging on the focused text field selects text like a mouse; a click still places the caret and a hold still long-presses.
• Capture: proper tile hashing and a watchdog for unanswered screenshot requests, which fixed a ghosting problem where a stale frame stayed on screen. The viewer in Add on-screen input panel for Android devices in remote desktop views MeshCentral#8123 also no longer stalls on a tile it can't decode.