An MCP tool for coding agents
Your agent can’t tap a real iPhone. Our testers can.
When your AI agent needs to know whether something actually works for a person on a real device, it asks Proofhand. Real testers run the steps and report back pass or fail, with screenshots, usually within the hour.
Pay per run. No subscription. Unused runs are refunded.
● I fixed the checkout layout. I can’t check it on a real iPhone, so I’ll ask Proofhand.
proofhand.create_check({ url: "staging.acme.dev/cart", devices: ["ios-safari", "android-chrome"], steps: 3, reward: $4 })
● Waiting for testers… (11 min)
iPhone: 1 failure found
✓ pass Add an item to the cart
✓ pass Tap Checkout
✗ fail Pay with the test card: “Pay button hidden behind the keyboard, had to close it.” 📎 2 screenshots
● Found it: the sticky footer ignores the visual viewport. Fixing, then approving the run.
Real devices
iPhone, Android, Mac and Windows browsers, plus TestFlight and Google Play betas.
Evidence, not vibes
Every failed step comes with a note and a screenshot or recording.
Built for agents
Structured JSON results, long-polling, webhooks and idempotent retries.
Only your apps
Checks can only target domains and betas you’ve proven you own.
How it works
- Step 1
Connect
Verify your staging domain, add funds, and run one command to give Claude Code or Cursor the Proofhand tools.
- Step 2
Ask
Mid-task, your agent posts a check: a URL, a few concrete steps, and which devices to try.
- Step 3
Fix
Testers report each step as pass, fail or blocked. Your agent reads the report, fixes the bug, and approves the run.
Pricing
You set the reward per run, from $2 to $100. We add 20% on top. A 3-step check on two phones at $4 each costs $9.60.
- Funds are reserved when you create a check
- Runs nobody claims are refunded automatically
- Approve or reject within 72 hours, or it approves itself
Test apps, get paid
Use the phone or laptop you already have. Pick a check, follow the steps, report what happened. You’re paid for the work, whether the app passes or fails.
- Work when you want, 5–20 minutes per check
- Cash out to your bank from $10
- Unfair rejections go to a person, not a bot
Questions
What can’t I use it for?
Anything that defeats someone else’s protections: CAPTCHAs, creating accounts on sites you don’t own, verification codes, reviews or ratings. Checks can only target domains and betas you’ve verified.
Are my staging credentials safe?
Test credentials are encrypted at rest and shown only to the one tester currently holding that run. Use staging accounts, never production ones.
What if a tester does a bad job?
Reject the run with a reason that references the steps. The funds stay reserved while the tester can dispute, and a person decides.
How fast is it?
It depends on reward and devices. Most checks are picked up within minutes during the day. Your agent can long-poll or use webhooks.