Your site has bugs.
An AI finds them first.
Point it at your live site. An AI agent clicks through it like a real user, signing up, filling forms, following your flows, and reports every broken thing with a screenshot. No scripts to write, no test suite to maintain.
What it finds
The things a real user would call broken
How the agent explores
It looks, decides, acts, and reports, one human-like step at a time, until it has covered your main flows.
It looks at the page
Every step, the agent takes a screenshot and reads the clickable things on the page, buttons, links, inputs.
It decides like a user
Claude vision picks the next human move: sign up, open the pricing page, submit the form, or flags something broken.
It does it, and watches
The agent clicks or types, watching for JS errors, 500s, dead clicks, and blank or crashed pages.
It reports with a screenshot
Every bug lands with a severity, the page, and a screenshot, plus the full trail of what it tried.
Everything it does
More than a smoke test
A curious agent that follows your actual flows, re-checks nightly, and tells you what newly broke.
Explores like a real user
Not a page-load smoke test, a curious agent that follows your actual flows and covers new ground each step.
Logs in and fills forms
Give it test credentials for the run (never stored) and it exercises signup, checkout, settings, multi-step flows.
Nightly runs and alerts
Schedule a suite to run daily or weekly. Results ping you in-app, over a webhook, or a subscribable RSS feed.
Regression diffs
Every run is compared to the last: what newly broke, what got fixed, what is still open. Three new issues since last night.
Shareable bug reports
One click mints a public report link, hand a client or a teammate a URL, not a file.
Parallel journeys
Run up to three exploration paths at once, signup, core features, settings, for faster, wider coverage.
Fix with an AI team
For apps built on MCPity, hand the findings to a Builder agent that proposes the fix as a draft PR you review.
Safe by default
Never deletes, cancels a subscription, or submits a real payment, a two-wall denylist, plus a per-run step cap.
What you get
A report, not a red X
Every finding comes with a severity, the page, a plain-English note, and a screenshot, plus how it compares to your last run.
Clicked the primary CTA, no navigation, no network request, no feedback.
Submitted the form with a blank password field; it advanced instead of validating.
“Careers” in the footer resolves to a 404 page.
Illustrative, your report shows the real bugs the agent found on your site.
What it catches
The bugs users hit
Your own AI key
The vision calls run on your Anthropic key, no markup. It never leaves the server.
1 credit per run
Flat and simple. A whole exploration, dozens of steps, screenshots, a report, is one credit.
Safe by default
Never deletes data, cancels a plan, or pays. Test credentials are used once and thrown away.
Questions
Good to know
Does it work on any website?
Any public URL. For pages behind a login, give it test credentials for the run, they are used once and never stored. It works on a site you built on MCPity or a totally external one.
Whose AI key does it use?
Yours. The vision calls run on your own Anthropic key, with no markup, and the key never leaves the server, the browser only ever gets instructions back. MCPity's only cost is browser time.
Is it safe to run on my live site?
Yes. Safe mode is on by default: it never deletes data, cancels a subscription, or submits a real payment. There is a destructive-action denylist as a second wall, plus a per-run step and time cap so it cannot run forever.
How is this different from a test suite?
You write nothing and maintain nothing. Instead of scripted assertions, a curious agent explores like a real user and surfaces bugs you did not think to test, then re-checks nightly and tells you what newly broke.
Can it fix the bugs too?
For apps built on MCPity (a connected repo plus Supabase), yes, hand the findings to a Builder agent that proposes the fix as a draft PR you review. For an external site with nothing connected, it tests but cannot fix.
What does it cost?
1 MCPity Credit per run, plus your own AI usage on your key. A whole exploration, dozens of steps, screenshots, and a report, is a single credit. No subscription required to run it.
Ready when you are
Find your bugs first
Paste your URL and let the agent explore. You get a report, every bug, every screenshot, in minutes.