Let a Coding Agent Run Your App on a Real Device
A coding agent can write a mobile feature but usually cannot see whether it works — it edits, you run, you describe what happened. The VibeView CLI closes that loop. It is a published npm package, so any agent that can run a shell command can upload a build, drive a live iOS or Android device, run a test and read the result as JSON. No plugin, no adapter, nothing to install on our side.
How it works
-
1. Install the CLI and authenticate once
Install globally and log in yourself — the browser login is the one step an agent should not do for you. After that, an API token in the environment lets the agent work unattended.
npm install -g vibeview vibeview login # for unattended agent runs export VIBEVIEW_API_TOKEN=... -
2. Give the agent a build to work against
Upload a build once and the agent can reuse it. Simulator builds for iOS (.app, or compressed .zip/.tar.gz/.tgz), an .apk for Android, up to 500 MB. list-devices and list-apps let the agent discover what is available instead of guessing.
vibeview upload-app path/to/YourApp.app --output json vibeview list-apps --output json vibeview list-devices --output json -
3. Ask for JSON, not human output
This is the flag that matters. Every command takes --output json, so the agent parses a result rather than scraping formatted text — and run-suite additionally emits junit if you want the same run readable by CI. Steps are plain-English descriptions, which is a format an agent writes well.
vibeview create-test "Login flow" \ --app abc123 \ --steps '[{"description":"Open app"},{"description":"Tap Login"},{"description":"Verify home screen"}]' \ --output json vibeview run-test <test-id> --output json -
4. Or let it watch its own changes live
For React Native work, vibeview dev connects the Metro bundler in the project to a cloud device, so a change the agent makes appears on a running device about a second after it saves. Pair that with a session the agent opens and you get an edit-run-look loop where you are reviewing rather than relaying.
vibeview dev --platform ios --app abc123 -
5. What is not built yet
There is no VibeView skill file and no MCP server, so an agent does not discover these commands on its own — you tell it, or you keep the invocations in your project notes. Everything above works today through the CLI alone. The same is true of Cursor and any other agent with terminal access; nothing here is specific to one tool.
Full setup instructions live in the documentation.
Add Claude Code to your pipeline
Start free and run your first check on a live device.