callstackAutomation & workflows
Agent Device
agent-device
Mobile app automation and verification for AI coding agents, enabling live interaction and testing across iOS, Android, HarmonyOS, and more.
- Frontend
- QA / Testing
- DevOps
- Automation
- Testing
- Monitoring / Observability
- Productivity
- CLI
- API/SDK
- Library/framework
- macOS
- Linux
- Windows
- Browser
- Self-hostable
- Runs locally
- Popularity
- 4.2k Stars
- GitHub stars
- Recent activity
- 8/22/2026
- Updated in the last 30 days
- License
- MIT
- Permissive
Why it matters
We look beyond stars: what problem it solves, whether it creates real utility, and what makes its approach worth noticing.
Problem
Addresses the lack of real-time app feedback and automated verification mechanisms for AI coding agents in mobile development.
Practical value
Provides a unified interface via CLI, MCP server, and Node.js API, supporting platforms like iOS, Android, and HarmonyOS with capabilities for inspection, interaction, and evidence capture.
Innovation / differentiation
Converts accessibility trees into token-efficient text snapshots for AI comprehension, moving beyond reliance on pure visual recognition to improve interaction accuracy.
Leverage potential
Integrates into existing AI development workflows, supporting local debugging, CI/CD automation, and remote device cloud connectivity, extending the execution scope of AI agents on mobile.
Why now
As AI coding agents become more prevalent in complex application development, automated verification tools tailored for mobile environments are seeing increased demand.
Community activity
In the last 90 days there were 367 new issues and 989 pull requests; the bounded issue/PR samples include 18 issue authors and 4 PR contributors, with at least 41 releases.
Maintainer responsiveness
The 100-issue window sample had a 99% close rate, and the 100-pull-request sample had a 95% merge rate. Maintainer-response observations covered 68% of that issue sample, with a 3% response rate and median first response of 9.1 hours.
Key highlights
- Inspect app state via accessibility snapshots
- Perform UI interactions like tapping, typing, and scrolling
- Capture screenshots, videos, logs, and performance data as evidence
Quick start
How it is installed, how hard it is, and where to start.
Where it runs
Runs locally
Difficulty
Medium — some setup needed
- 01Install the agent-device tool globally using npm.
- 02Run the doctor command to verify that your environment meets the requirements for target platforms.
- 03Start an app session via the CLI and use the snapshot command to retrieve the current UI state.
- 04Perform interactions such as press or fill using the returned refs.
Best for
- Teams that want data on their own servers
- Developers who want to try it on their machine
More about it
agent-device provides a live feedback loop for AI coding agents to interact with mobile applications. Through a CLI, MCP server, or Node.js API, it allows agents to inspect, control, and verify apps across iOS, Android, HarmonyOS, web, and desktop platforms.
By leveraging accessibility trees rather than relying solely on visual reasoning, the tool ensures reliable interactions. Developers can record agent workflows as scripts for local debugging or integrate them into CI/CD pipelines for automated mobile quality assurance.
Sources
Each field shows its status and source — expand to review.
14 · Expand
Sources
Each field shows its status and source — expand to review.
capability tags
Verifiedautomation, testing, monitoring, productivity
Source: admin_cms · cms editor · 8/22/2026
editor note
Verified{"en":"By providing an MCP server and Node.js API, this project enables AI agents to operate simulators or physical devices, creating an 'inspect-act-verify' loop. It is particularly useful for workflows requiring automated mobile UI verification or bug reproduction.","zh":"该项目通过提供 MCP 服务器和 Node.js API，让 AI 代理能够直接操作模拟器或真机，实现“检查-行动-验证”的闭环。它特别适合需要自动化验证移动端 UI 变更或复现 Bug 的开发工作流。"}
Source: admin_cms · cms editor · 8/22/2026
how to use
Verified{"steps":[{"en":"Install the agent-device tool globally using npm.","zh":"使用 npm 全局安装 agent-device 工具。"},{"en":"Run the doctor command to verify that your environment meets the requirements for target platforms.","zh":"运行 doctor 命令检查环境配置是否满足目标平台要求。"},{"en":"Start an app session via the CLI and use the snapshot command to retrieve the current UI state.","zh":"通过 CLI 启动应用会话，并使用 snapshot 命令获取当前界面状态。"},{"en":"Perform interactions such as press or fill using the returned refs.","zh":"根据返回的引用（ref）执行交互操作，如 press 或 fill。"}],"installAt":"local","difficulty":"medium"}
Source: admin_cms · cms editor · 8/22/2026
intro
Verified{"en":"agent-device provides a live feedback loop for AI coding agents to interact with mobile applications. Through a CLI, MCP server, or Node.js API, it allows agents to inspect, control, and verify apps across iOS, Android, HarmonyOS, web, and desktop platforms.\n\nBy leveraging accessibility trees rather than relying solely on visual reasoning, the tool ensures reliable interactions. Developers can record agent workflows as scripts for local debugging or integrate them into CI/CD pipelines for automated mobile quality assurance.","zh":"agent-device 为 AI 编程助手提供移动端应用的实时反馈回路。它通过命令行工具、MCP 服务器或 Node.js API，让 AI 能够直接在 iOS、Android、HarmonyOS、Web 及桌面端应用中进行检查、控制和验证。\n\n该工具通过读取无障碍树（Accessibility Tree）而非仅依赖视觉识别，确保了操作的准确性。开发者可以将 AI 的操作过程保存为脚本，用于本地调试或集成到 CI/CD 流水线中，从而实现自动化的移动端质量保证。"}
Source: admin_cms · cms editor · 8/22/2026
Latest release
Verifiedv0.20.9
Source: GitHub API · latest_release=v0.20.9 · 8/22/2026
License
VerifiedMIT
Source: GitHub API · license.spdx_id=MIT · 8/22/2026
needs api key
VerifiedNo
Source: admin_cms · cms editor · 8/22/2026
One-liner
Verified{"en":"Mobile app automation and verification for AI coding agents, enabling live interaction and testing across iOS, Android, HarmonyOS, and more.","zh":"为 AI 编程助手提供移动端应用自动化与验证能力，支持 iOS、Android、HarmonyOS 等多平台的实时交互与测试。"}
Source: admin_cms · cms editor · 8/22/2026
Platforms
Verifiedmacos, linux, windows, browser
Source: admin_cms · cms editor · 8/22/2026
Category hint
Inferred from materialsai-apps
Source: Project README · hint=ai-apps · 8/17/2026
product forms
Verifiedcli, api_sdk, library_framework
Source: admin_cms · cms editor · 8/22/2026
role tags
Verifiedfrontend, qa, devops
Source: admin_cms · cms editor · 8/22/2026
supports local
VerifiedYes
Source: admin_cms · cms editor · 8/22/2026
supports self host
VerifiedYes
Source: admin_cms · cms editor · 8/22/2026
Related projects
Other verified projects matched by category, capabilities, and intended roles.
woodpecker
A self-hosted, plugin-extensible CI/CD engine with SQLite available by default for the server and a focused pipeline workflow.
Rundeck
A self-hosted runbook automation service with web, CLI, and API interfaces for turning operational procedures into repeatable jobs.
dsh-chat-import
Import conversation histories from 12 AI agents like Claude Code, Cursor, and ChatGPT into DeepSeek Harness with full-fidelity resume and sync capabilities.
hh-ai-agent
Job-search agent that finds HH.ru vacancies, evaluates them with an LLM, generates cover letters, and sends suitable matches to Telegram.
marketing-studio
An agent-driven marketing studio for Claude Code that generates product demos, logo reveals, launch videos, and social assets from a single command.
wechatauto-replica
An automation library for WeChat 4.x Windows client, enabling message reading, history export, and automated sending via local database decryption and hybrid UI drivers.