callstack/agent-device

Mobile app automation and verification for AI coding agents. CLI, MCP server, and typed Node.js API for iOS, Android, HarmonyOS, TV, web, macOS, and Linux.

View on GitHub ↗Jump to charts ↓Open shareable report

Summary Information

Updated 16 minutes ago
Added to GitGenius on September 15th, 2026
Created on January 30th, 2026
Open Issues & Pull Requests: 89 (-3)
GitHub issues: Enabled
Number of forks: 298
Total Stargazers: 4,587 (+4)
Total Subscribers: 8 (+0)

Repository Insights (GitGenius)

Median issue/PR response: 7.0 hours
Mean response time: 3.7 days
90th percentile: 5.1 days
Tracked items: 683

Most active contributors

Sign in to see contributor activity.

How this project is maintained

Around half of the issues opened in the past year never receive a reply. Only 16% of open issues come from outside the core team — the tracker reads mainly as internal planning. Almost all tracked open issues have seen activity in the last three months. Only 14% of issues opened in the past year have been closed. Three people close 99% of everything that gets resolved.

Charts & Analytics

Fetching additional details & charts...

Issue Activity (beta)

Open issues: 69
New in 7 days: 85
Closed in 7 days: 63
Avg open age: 30 days
Stale 30+ days: 14
Stale 90+ days: 3

Recent activity

Opened in 7 days: 82
Closed in 7 days: 60
Comments in 7 days: 0
Events in 7 days: 0

Top labels

  • ready-for-agent (254)
  • enhancement (129)
  • bug (102)
  • refactor (92)
  • needs-triage (72)
  • in-progress (29)
  • backlog (22)
  • ready-for-human (15)

Detailed Description

agent-device is a mobile app automation and verification tool that gives AI coding agents a live feedback loop through a CLI, MCP server, or typed Node.js API.

The tool solves the problem of agents lacking real-time verification of their changes in running applications. Instead of reasoning only from screenshots, agents work with token-efficient accessibility snapshots extracted from the app's accessibility tree, act through element references and selectors, and capture evidence for review. The approach supports iOS, Android, and HarmonyOS on simulators, emulators, and physical devices, alongside tvOS, Android TV, Amazon Vega OS TV, web, macOS, and Linux. It coordinates device access across parallel agent worktrees and integrates with remote device clouds.

The tool suits teams building or using AI coding agents that need to verify mobile app changes. It works with Claude Code, Codex, Cursor, Windsurf, Cline, Goose, and any agent capable of running a CLI or connecting over MCP. Developers can also build custom agents using the AI SDK or Eve framework with the tool as the runtime. The accessibility-tree approach makes agent runs more reliable when apps have clear labels, roles, and test IDs, though screenshots and video remain available as evidence or fallback when accessibility data is sparse. The tool requires Node.js 22.12 or newer, with Node.js 24 or newer needed for web automation.

Development activity shows consistent engagement with the codebase through regular commits addressing both features and maintenance. The project maintains comprehensive documentation that stays synchronized with the installed version, helping users debug, replay, and profile agent behavior. The maintainers provide per-client setup guidance and actively support integration with multiple agent platforms, indicating ongoing investment in the tool's usability across different deployment contexts.