Mobile Agent Harness

io.github.Aelindrav0.5.0Updated Oct 4, 2026

Android device automation for AI agents: MCP server, CLI, and plugin runtime over adb.

VerifiedSTDIODesktop onlyDeveloper ToolsBrowser Automation

Overview

AI-generated overview

Lets an AI assistant observe and control an Android device over adb, using accessibility-tree selectors, screenshots, OCR, and app flows.

What it does
An MCP server, CLI, and plugin runtime that drives Android devices through a self-healing uiautomator2 bridge, with no device-side app needed. Observation tools include ui.snapshot, ui.dump, ui.find, vision.screenshot, vision.ocr, and vision.diff; action tools include ui.click, ui.set_text, ui.scroll, input.tap, and input.swipe. It also exposes app lifecycle tools (app.launch, app.probe, app.wait_idle), shell and data tools (shell.run, state.prefs, state.db, file.push, file.pull, net.http), event capture, and knowledge/task helpers. Selectors resolve against the live accessibility tree rather than coordinates, and tools return an ok/error envelope.
When to use it
Use it when an assistant needs to inspect or operate a real Android device or emulator: reproducing UI flows, reading on-screen state, running app-level checks, or building per-app automation harnesses for development and testing on your own device.
Requirements
Runs locally as a Python process (Python 3.9+) with adb on PATH and USB debugging enabled on the target device. Device serial is set via AGENT_SERIAL_DEFAULT, AGENT_SERIAL_USB, or --serial/ANDROID_SERIAL. Optional vision features use MAH_FRAME_STREAM, MAH_VISION_MODE, and MAH_VLM_URL/MAH_VLM_MODEL; the OCR extras require installing the ocr extra.
Before you install
The server can act on a device: clicking, typing, scrolling, launching and stopping apps, running shell commands, and pushing or pulling files, so mistakes can change or delete device data. Root support is declarative and only active if the device already has su, gated by a command allowlist. Messaging, banking, and e-commerce apps use anti-automation risk control and the adb tier is unreliable there. If MAH_VLM_URL points at an external vision endpoint, screen data is sent to that third party.

Installation

In SourceWeft

  1. Open Mobile Agent Harness in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

mobile-agent-harness

Android device automation for AI agents. MCP server, CLI, and a hot-reloadable plugin runtime on top of a self-healing uiautomator2 bridge. No device-side app required.

中文文档:README.zh-CN.md

This is the base runtime only. Plugins live in the community catalog: awesome-mobile-agent-harness-plugin. App business knowledge for AI lives in app-knowledge-packs, which this repo's knowledge/ loader reads directly.

Install

bash
git clone https://github.com/Aelindra/mobile-agent-harness
cd mobile-agent-harness
pip install -e .          # add ".[ocr]" for vision.ocr / ui2.*

Requires Python 3.9+ and adb on PATH with USB debugging enabled.

Quick start

  1. Point the bridge at your device:
bash
export AGENT_SERIAL_DEFAULT=192.168.1.23:5555   # or "emulator-5555"
  1. List the tool surface (works without a device) and run the offline tests:
bash
python -m core.cli tools
python tests/test_runtime.py
  1. Call a tool:
bash
python -m core.cli call ui.snapshot
python -m core.cli call ui.click --args '{"selector":"text@^Network & internet$","pkg":"com.android.settings"}'
  1. Attach to any MCP client:
json
{
  "mcpServers": {
    "phone": {
      "command": "python",
      "args": ["-m", "core.cli", "mcp"],
      "cwd": "/path/to/mobile-agent-harness",
      "env": { "AGENT_SERIAL_DEFAULT": "192.168.1.23:5555" }
    }
  }
}

All tools return a {"ok": bool, "error"?, ...} envelope. ok=false is a business result (e.g. error_type=selector_not_found) — adapt instead of retrying.

Tools

GroupTools
Observationui.snapshot ui.dump ui.find ui2.state ui2.screen_evidence ui2.timeline ui2.find_templates vision.screenshot vision.ocr vision.diff
Actionui.click ui.click_handle ui.text_handle ui.hold_read ui.set_text ui.scroll ui.back ui.home input.tap input.swipe input.key
Appapp.launch app.current app.probe app.list app.stop app.wait_idle
Shell & datashell.run state.prefs state.db file.push file.pull net.http
Eventsevents.tail events.wait events.capture events.record_start events.record_stop
Knowledge & tasksknowledge.list knowledge.guide ui2.check_states ui2.orient task.begin task.note task.end template.add sys.capabilities

ui.snapshot returns a compact accessibility-tree listing with element handles (e0, e1, ...); ui.click_handle acts on a handle with drift detection. ui2.state fuses the accessibility tree with OCR for canvas-drawn UIs. ui2.screen_evidence collects measurable screen features (overlay level, motion, color, layout density) without interpreting them. app.probe returns deterministic side signals (package, version, orientation); ui2.timeline captures N frames of lightweight evidence in time order; ui2.orient probes context, routes to matching knowledge packs, and evaluates their state discriminators — returning matched states or ranked hypotheses with the missing evidence listed.

Selectors

Selectors resolve against the live accessibility tree. Coordinates are not used.

python
# ui.click selector examples
"id@d98 && [email protected]"       # resource-id, short form completed with pkg@
"text@^Settings$ && [email protected]"
"[email protected] && [clickable=true]"
"desc@^Search$"                        # content-desc regex

Harnesses and plugins

A harness file declares per-app flows as selector steps with pre/postconditions; distill drafts one by exploring an app, lint checks selector quality, and failed postconditions mark tools stale for re-distillation. See harnesses/com.android.settings/harness.json5 for the built-in example.

Plugins drop into plugins/ and register tools or capability providers at load time; registrations are reversible and hot-reloaded on file change. See the Plugin development guide in README.zh-CN.md and knowledge/README.md for the knowledge-pack format.

For business context on apps where model priors are unreliable, point agents at app-knowledge-packs — a community catalog of structural app surveys in the open Agent Skills format, split per feature module for complex apps. Mobile scenarios in particular are private-domain and underrepresented in training data, so agents benefit from loading the relevant survey before operating an unfamiliar app. The local knowledge/ directory remains this repo's runtime layer (packs, icon templates, state discriminators).

Configuration

VariableDescriptionExample
AGENT_SERIAL_DEFAULTDevice serial used when --serial/ANDROID_SERIAL absent192.168.1.23:5555
AGENT_SERIAL_USBUSB serial fallback0123456789abcdef
MAH_FRAME_STREAM1 enables the H.264 frame stream for observation tools1
MAH_VISION_MODEVision enhancement: off / client / endpointclient
MAH_VLM_URL / MAH_VLM_MODELExternal vision endpoint (OpenAI-compatible)http://127.0.0.1:11434/v1/chat/completions

How it compares

  • vs mobile-mcp / agent-device: they are generic device toolsets; this project adds the layer above transport — per-app knowledge as files (harnesses), a hot-reloadable plugin runtime with capability seams, and an explicit adb/root privilege model.
  • vs droidrun: droidrun requires a device-side accessibility app; this project is zero-install over adb.
  • Root support is declarative: if the device already has su, the shell seam gains a root provider gated by a command allowlist. This project does not provide root.

Known limitations

Messaging, banking, and e-commerce apps run active anti-automation risk control; the adb tier is unreliable on them. Target use is development, testing, and your own device workflows.

Contributing

Plugins are contributed to awesome-mobile-agent-harness-plugin; app business knowledge to app-knowledge-packs — see their contributing guides.

License

MIT.

Source: README.md at commit a05ec97

Tools

0
Tool metadata has not been indexed yet.

Version history

1
  1. v0.5.0LatestOct 4, 2026