Scan a live site before any traffic arrives: what it does and how to use it
The UserProof scan opens your live site in a headless browser, clicks what looks clickable on up to 8 pages, checks up to 80 internal links and returns the one worst defect it can reproduce 3 of 3 times. It takes 1 to 3 minutes, never submits forms, and is free up to 10 a day per site.
This page is for founders who build with a coding agent (Claude Code, Codex or Cursor) and have shipped a site nobody has visited yet. How common are broken first screens? In a related run on October 28, 2026, 122 of 378 Y Combinator startup homepages (32%) showed no sign-up or demo button on the first phone screen (YC homepage report). The free homepage grader scores your own homepage on the same 10 checks before the scan runs. That run used the same kind of test, a headless Chromium that clicks and retries in fresh pages. It also had a main-button check that scan_site only runs when you have set a goal, and it opened up to 3 pages per site where the scan opens up to 8.
Jump to: scan or wait · what it clicks and checks · what it reports · run it · limits · price · sources · FAQ
Scan or wait for real visitors?
Scan first if nobody has visited yet, and wait for visitors for anything a person has to feel. Each row says what finds it.
| What you want to know | The scan (day one) | Real visitors (after traffic) |
|---|---|---|
| A button or link that does nothing | Yes: dead click, no effect in 5 seconds, 3 of 3 tries | Yes, with how many visitors hit it |
| A link that returns 404 or 410 | Yes, up to 80 internal links | Only if a visitor clicks it |
| A script error on load or on click | Yes, from your own scripts | Yes, on the devices visitors use |
| A response slower than 2 seconds | Yes, on one desktop browser | Yes, on real networks and phones |
| Rage clicks and where people drop off | No | Yes |
| Pages behind a login | No | Yes |
| Whether it costs you sign-ups | No: a scan result is labelled a scan, not visitor behaviour | Yes, once there is enough traffic |
| Cost | Free, 10 a day per site | Plans from $0, priced on recorded visits |
What does the scan click and check?
It loads up to 8 pages of your own domain in headless Chromium at 1366 by 860 pixels, starting from the home page or the URL you give, and tests five things. We read this from the scanner source on November 9, 2026.
- Clicks. On each page it picks up to 20 elements that look clickable (buttons, tabs, menu items, summaries, links with no real target, and plain elements with a pointer cursor). They are ranked by call-to-action wording, being a real control and sitting on the first screen. It moves the mouse onto each, clicks, and watches for any visible effect for 5 seconds: a page change, a DOM change, a scroll, a popup, a dialog, a download or a blocked request.
- Dead, slow and error clicks. No effect in 5 seconds is a dead click. An effect after more than 2 seconds is a slow response. A script error from your own code within 1.5 seconds of the click is an error click.
- Links. It requests up to 80 internal links with a plain GET and reports any that return 404 or 410 twice in a row.
- Load problems. It reports scripts from your own domain that throw when the page loads, and scripts, stylesheets and API calls on your domain that return a status of 400 or higher.
- Repeat before report. A failed click is retried in two fresh pages. It is reported only if all 3 tries fail the same way.
It is read-only. Every non-GET request is blocked, analytics beacons are blocked, off-domain links are not followed, and popups and dialogs are closed. If you have set a goal for the site, it also walks that goal once on a phone and a desktop and adds what it finds. For what counts as a dead or rage click in recorded visits, see how UserProof finds dead and rage clicks.
What does the scan report, and in what shape?
It reports one worst defect in plain text, plus a count and list of the rest. Defects rank by kind: an error click scores 100, a dead click 80, a broken link 60, a slow response 50, a load error 30 and a failed request 25. A call-to-action wording, a real control or a first-screen position adds points, and the same defect on several pages merges into one.
The top defect comes back as a line like SCAN RESULT for yoursite.com (found by our scan, not yet by visitors): "Get started" looks clickable but nothing happens when clicked (3 of 3 tries, watched 5 s), then the pages it was on, the element tag and CSS selector, and a defect id. The structured reply also carries the element's HTML, the evidence from each try, a short list of the other defects (id, kind, page, text) and stats: pages, links checked, defects and duration. If nothing is wrong it says how many pages it scanned and links it checked, and that real visitor data will show what a scan cannot.
How do I run it from my coding agent?
Connect UserProof to your agent, then ask it to call scan_site. The server setup is on the MCP server page, so these steps start after it.
- Sign in and pick the site. Run the setup once (Claude Code steps);
select_sitesaves which site this project uses. Done looks like the agent naming your domain as the active site. - Give the agent the prompt.
Run scan_site on my live site and tell me the worst defect. Do not change code yet.To start somewhere other than the home page, add the URL. Done looks like a reply beginningSCAN RESULT for(orScannedfollowed by a page count when the site is clean) after 1 to 3 minutes. The tool waits for you; it polls for up to 6 minutes. - Have the agent find the code. The reply names a locator command,
userproof-locate, that searches your repo for the element by its text, classes, id and link. Done looks like a file and line. - Fix it, deploy, and re-test. After the fix is live, the agent calls
mark_fix_appliedwith the defect id, thencheck_fixrepeats the scan test 3 times in fresh pages. Done looks like a verdict of fixed. The other verdicts are still broken, element gone and inconclusive. More on that in check whether a shipped fix worked.
Want to test one page by hand first? The free dead click finder runs in your own browser and uses the same 5-second rule.
What can the scan not find?
It cannot find anything that needs a person, a login or a different device, and it reports one worst defect first, not a full audit.
- Nothing behind a login. It is a logged-out visitor and never signs in.
- No typing. It does not fill forms, and it skips a submit button whose fields are empty, so a broken form is not judged.
- One browser size. Pages load at 1366 by 860 pixels. Phone layouts are checked only in the goal walk, and only when a goal is set.
- A slice of the site. Up to 8 pages, 20 clicks a page and 80 links. Elements already in an active state, undo and redo buttons, hidden elements and ones covered by another element are skipped.
- Clicks it cannot see. An effect it does not observe reads as a dead click, so check a hover-only menu or an in-page action before changing it.
- Your domain only. A URL on any other domain is refused with a 400 error.
- No visitors, no conversion. It cannot tell you what costs you sign-ups; a defect becomes a visitor finding only after enough recorded sessions.
- Time. It stops starting new clicks after 150 seconds, so large sites finish partly scanned. Scans run one at a time on our servers, so a busy day can add a wait.
The limit is 10 scans in any 24 hours per site, and the eleventh is refused.
What does the scan cost?
It is free: up to 10 scans a day per site, on Free, Pro and Scale. The scan code has no plan check, and the pricing page lists site scans as included in Free.
| Plan | Price | Recorded visits a month | Sites | Scans |
|---|---|---|---|---|
| Free | $0 | 5,000 | 3 | 10 a day per site |
| Pro | $29 a month | 50,000 | Unlimited | 10 a day per site |
| Scale | $99 a month | 250,000 | Unlimited | 10 a day per site |
Analyses and fix checks are unlimited on every plan. A scan is not a recorded visit and does not count toward the monthly limit, and the scan blocks UserProof's own beacons so it records nothing.
How a defect becomes a line of code, with the locator’s measured accuracy, is on the file and line page.
Sources, read November 9, 2026
- UserProof scanner and API source in the GitHub repo (click rules, page and link caps, read-only rules, report text, the 10 a day limit and the 429 answer), read November 9, 2026
- UserProof company page (the capability: a free scan of the live site finds broken elements on day one, before real visitors arrive), read November 9, 2026
- UserProof pricing page (site scans free up to 10 a day per site, plan prices), read November 9, 2026
- UserProof YC homepage report (378 sites, 122 with no sign-up or demo button on the first phone screen, method), checked October 28, 2026, read November 9, 2026
Comparing tools rather than running a scan? Session replay tools for coding agents, ranked covers what each one does.
FAQ
Does the scan change or submit anything on my site?
No. Every request other than a page load is blocked, so no form posts and no checkouts go out. Analytics beacons are blocked too, so the scan does not touch your own numbers. It skips submit buttons whose fields are empty, closes popups, dismisses dialogs and does not follow links to other domains.
How many scans can I run?
Free up to 10 a day per site, on every plan. The eleventh start inside 24 hours is refused with a 429 error that says the scan limit is reached. If a scan on that site is already running, a new request returns the running one instead of starting another. Analyses and fix checks are unlimited.
Is the scan the same as the dead click finder?
No. The dead click finder runs in your own browser on the page you open and judges the clicks you make. The scan runs from our servers in a fresh headless browser, loads up to 8 pages by following your links, picks the clickable elements itself and repeats each failure 3 times before it reports it.
Can it scan a site behind a login or a local dev server?
No. It loads the site as a logged-out visitor from our servers, so pages behind a sign-in are not seen and the site must be reachable on the public web. It also runs only on the domain registered on your account and answers 400 for any other domain.