Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.tapkit.ai/llms.txt

Use this file to discover all available pages before exploring further.

Demo video coming soon

​
What this does

The agent runs end-to-end test flows on your iPhone against real production apps — or your own app in development. It follows a test script, verifies expected behavior at each step, and reports pass/fail results. Because it runs on a real device, you’re testing against the actual app experience, not a simulated version.

​
How we built it

We wrote test prompts as natural language scripts describing the steps and expected outcomes. For example: “Open the app, tap Sign In, enter test credentials, verify the home screen loads, tap the profile icon, verify the profile name matches.” The agent executes each step, takes a screenshot, verifies the expected state, and moves on. If something doesn’t match expectations, it reports what it saw instead of what was expected. This approach works particularly well for:
  • Smoke testing after a release — quickly verify critical flows work
  • Cross-app testing — test interactions between your app and other apps (deeplinks, share sheets)
  • Third-party app testing — verify your app works correctly on real iOS, not just simulators
  • Regression testing — run the same test script across app versions
The biggest advantage over traditional automation frameworks: no test IDs, no XCUITest setup, no simulator configuration. The agent just looks at the screen and acts.

​
Try it yourself

Paste this into the TapKit Mac app agent or Claude with TapKit connected:
Open Safari. Navigate to example.com. Verify the page loads and contains the heading “Example Domain.” Tap the “More information…” link. Verify the new page loads. Take a screenshot and report whether each step passed.
Works with: TapKit Mac app, Claude.ai, Claude Code, any MCP client.

​
Tips and things to know

  • Write test prompts like scripts — describe each step and what you expect to see. “Tap X, verify Y appears, tap Z” is the pattern.
  • The agent takes verification screenshots — it compares what it sees to what you described. Include specific text or UI elements to look for.
  • Works on any app — unlike XCUITest, the agent doesn’t need accessibility IDs or test hooks. If a human can see it on screen, the agent can verify it.
  • Combine with the API for automation — use the Sessions API to run tests programmatically on a schedule.
  • Real devices catch real bugs — simulators miss issues with performance, network conditions, and hardware-specific behavior. Testing on a real phone finds the bugs your users will hit.
  • Great for manual test replacement — if you have a QA checklist that someone runs through manually, the agent can follow the same steps.