agent-device: verify mobile apps from your coding agent
Added New⧉ Copyable prompt below
Our verdict
The missing feedback loop when an agent writes mobile UI: it can open the app, read an accessibility snapshot, tap and fill by ref, then attach screenshots and logs. MIT-licensed from Callstack, with the same runtime behind the CLI, MCP and typed client. Needs Node.js 22.12+ (24+ for web), and a simulator, emulator or device; run agent-device doctor yourself before handing it to an agent.
- Who it's for
- Mobile and React Native teams using Claude Code, Codex, Cursor or Goose who need the agent to check its own UI work
- Leverage
- Agent-verified mobile changes with reviewable evidence
- Effort
- Medium
- Date added
What it does
agent-device keeps device state in sessions and drives apps through accessibility trees (not screenshots alone). It supports iOS, Android and HarmonyOS (simulators, emulators and devices), plus TV, web, macOS and Linux targets, with depth varying by platform. Agents act via refs and selectors, capture screenshots, video, logs and traces, save .ad replay scripts for CI, and can export Maestro YAML. agent-device mcp starts a stdio MCP server over the same command path as the CLI.
How builders use it
- Install with npm install -g agent-device@latest, then run agent-device doctor and agent-device help workflow.
- Add the MCP server to your client with command agent-device and args ["mcp"].
- Ask the agent to open the app, take an interactive snapshot, make a change, and attach screenshots.
- Save a working flow as a .ad script and replay it in CI.
What to expect
Pull requests with screenshots and logs from a real app session, not only a git diff.
Copy this: Install + MCP + first session
npm install -g agent-device@latest
agent-device doctor
agent-device help workflow
# MCP client config:
# { "mcpServers": { "agent-device": { "command": "agent-device", "args": ["mcp"] } } }
# Example CLI loop (iOS Contacts):
agent-device open Contacts --platform ios
agent-device snapshot -i
agent-device press @e2 --settle
agent-device screenshot ./evidence.png
agent-device close
Ask your agent:
Implement <change>, run it on the iOS simulator and Android emulator with agent-device, and attach screenshots plus any crash logs.Related discoveries
- WorkflowVibe-Coding & Prompt-to-App
Let your agent click through your app with Playwright MCP
Give the agent a real browser so it can check its own UI work.
Effort: Low⧉ Prompt#MCP#Open source
- ToolSmall Tools, Repos & Plugins
Repomix: pack a repo into one AI-ready file
One command turns a local or remote repo into a single file you can hand to any LLM, with token counts.
NewEffort: Low⧉ Prompt#GitHub repo#Plugin / extension
- ToolSmall Tools, Repos & Plugins
Serena: IDE-style code tools for your agent over MCP
An MCP server that gives coding agents symbol-level search, editing and refactoring backed by language servers.
NewEffort: Medium⧉ Prompt#GitHub repo#Plugin / extension
- ToolSmall Tools, Repos & Plugins
SEO Skill: Search Console facts for your coding agent
A free Apache-2.0 CLI and MCP server that answers SEO questions from your own Search Console, analytics and crawl data.
NewEffort: Low⧉ Prompt#Plugin / extension#Local / BYO model
Want more like this?
Practical discoveries, without the noise.