ARTEMIS
Turns natural-language instructions into Android automation: end-to-end workflows, logs, and MCP integration with AI coding assistants.
About
ARTEMIS is Google's open-source project for driving real Android devices from natural-language instructions. It runs a reactive observe-and-act loop — typically 3 to 5 seconds per step — with multimodal targeting that uses element indices when available and falls back to coordinate or visual locating for custom interfaces.
It ships as a native MCP server, so Antigravity, Claude Code, and Windsurf can drive test devices and pull Logcat output and screenshots directly into the IDE, with a one-click startup script that auto-installs ADB, scrcpy, FFmpeg, and Python dependencies. Google reports 99%+ task completion on its AndroidWorld benchmark of 100+ multi-step tasks — a provider-reported figure worth treating as a directional signal, not a guarantee.
What it can do
- Natural-language cross-app automation on real Android devices
- Reactive observe-and-act loop at roughly 3–5s per step
- Native MCP server for IDE diagnostics (Logcat, screenshots)
- Long-running exploratory and stability test support with blocked-action recovery
- One-click toolchain install and agent-rule setup for major AI IDEs
Setup
- Connect an Android device with USB debugging enabled (or an emulator)
- Run the one-click startup script — it installs ADB, scrcpy, FFmpeg, and Python deps
- Let it mount the global MCP server and agent rules into your AI IDE (Antigravity, Cursor, Claude Code, Codex, Windsurf, and others)
- Describe the workflow or test scenario in natural language
Prerequisites
Python 3.12+; an Android device with USB debugging or an emulator; a multimodal model (Gemini, Claude, GPT-4o, or Qwen-VL class).
Use cases
Works with
Trust score — 5/10
Scored 8/10 (METRICS tier, metrics-v1). Evidence: Google org, Apache-2.0, ~12k stars, active (pushed Oct 1 2026), Discord community, published docs with a real demo and workflow showcase. Docked for: (1) the 99%+ AndroidWorld completion figure is provider-reported on Google's own benchmark — no independent run; (2) real-device automation is inherently flaky across devices and OS versions, so mileage will vary. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · 12,045 GitHub stars; 1,299 forks; GitHub daily trending
- · pushed 2026-10-01; Google org; docs + Discord community + live demo
- · manual review: 99%+ AndroidWorld claim is provider-reported on own benchmark; real-device flakiness varies by device/OS
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.