You shipped it yourself — often with an AI coding agent — and there's no one to check it. Point Prufa at your URL. A real browser walks the site and reports what's actually broken.
No signup. No card. Public pages only, read-only.
We audited 49 Show HN launches in June 2026 — 78% had a critical bug live on day one. See the data →
You're one person. You wrote the code (or your AI agent did), you wrote the copy, you set up the deploy. There is no QA function — there's you, at midnight, clicking through your own signup form for the fourth time and hoping you didn't miss anything. The usual advice is "hire a contractor at $25–45/hr" or "ask a friend to poke at it." Neither is a process you can run on every deploy, and neither tells you whether your analytics actually fired or whether a link 404s on a page you never look at.
Prufa is an AI QA engineer that does the smoke test for you. The architecture is deliberately split: the LLM navigates and drives a real browser; plain code records the traffic and verifies the result against a spec. The model never judges whether your tracking is correct — it captures the network beacons and a deterministic check asserts them against what should have happened. Same input, same verdict. That's the whole point: you get a result you can trust without a human in the loop, and without learning a test framework.
It walks your public pages, captures every network request, console error, cookie, and response code, and grades the evidence into six plain-language sections: Works, Fast, Found, Measured, Accessible, Compliant. No 0–100 vanity score — just A–F grades and findings you can act on.
Three steps, no signup. Paste a URL and watch the run stream in live.
No signup, no card. The audit endpoint is anonymous — you hand it a public URL and it queues a run.
It loads your pages the way a visitor would and captures the actual network traffic, console output, cookies, and status codes — not a static crawl.
Deterministic checks grade the captured evidence. Machine-verified facts ship as VERIFIED findings; opinions ship separately as ADVISORY and never count against your grade. Results stream in live, then you get a shareable report.
If you build with Claude Code, Cursor, Cline, or Continue, you can wire Prufa in as an MCP server and have your agent verify what it just built — the same free audit, called as a tool. Install the open-source package and add one block to your MCP config:
Then your agent can call prufa_run_audit({url}) and read
the report back with prufa_get_report.
Both work anonymously — no key, no card. A
PRUFA_API_TOKEN is only needed later, for persistent
monitoring tiers. The package is open source (Apache-2.0,
prufa-dev/prufa-mcp).
Honest note: pip install prufa-mcp is the supported install
today. If you see npx/uvx invocations floating
around, those aren't published yet — use pip.
{ "mcpServers": { "prufa": { "command": "prufa-mcp" } } }Verified findings are machine-checked facts — these set your grade. The advisory block is opinion, never phrased as "broken", and never counts against the grade. The two tiers never interleave.
Your launch traffic is invisible — you can't see who signed up or where they dropped. This was the single most common critical in our 49-launch audit (38 of 49 sites).
Dead links break trust on the exact pages you want converting. 14 of the 49 launches shipped at least one.
Search engines may index the wrong variant of your page. 24 of the 49 launches had no canonical.
A first-time visitor may not understand the product from the fold alone — worth a second look. Advisory, not a verdict.
The straight answers to what solo founders ask before they paste a URL.
The free audit loads your public pages and records traffic read-only. It doesn't submit forms, doesn't make purchases, and doesn't mutate state. A core product invariant is that monitoring never pollutes your customer analytics. Honest scope: the free audit walks public pages — it doesn't log into authenticated flows. Testing a real signup or checkout flow is a separate, opt-in feature where you confirm the steps first.
No. A crawler reads HTML; Lighthouse runs lab metrics. Prufa runs a real browser through the page and asserts captured evidence against a spec — it checks whether your analytics beacons actually fired, not just whether a tag is present in the source. Where Lighthouse genuinely wins: deep performance lab diagnostics and a single throttled score. Prufa is about “does it work,” not “what's my Lighthouse number.”
Yes. Findings are plain-English headlines with A–F section grades and no 0–100 composite score (we deliberately don't fake that precision). The evidence line under each verified finding is there if you want to dig in, or hand to your AI agent to fix.
The questions solo founders actually ask before running their first audit.
Run an automated smoke test on every deploy instead of hiring. Point a tool like Prufa at your live URL and a real browser walks the site, capturing network traffic, console errors, broken links, and layout breakage, then grades the evidence. The free audit needs no signup or card and streams results live, so it fits a one-person workflow where QA would otherwise just not happen.
Solo founders typically face $25–45/hr for a freelance QA contractor, which isn't something you can run on every deploy. Prufa's audit is free — $0, no card — for a one-shot run on public pages, and paid monitoring starts at $29/mo if you want it on a schedule. See our breakdown of what a QA engineer actually costs.
At minimum, test the things that quietly fail — the stuff no one was assigned to check: confirm your signup and core money flow complete, check that analytics events actually fire, look for broken links and console errors, and open the site on a phone. In our June 2026 audit of 49 Show HN launches, no analytics firing was the most common critical (38 of 49); 14 shipped at least one broken link and 22 set cookies without the Secure flag. Our pre-launch checklist for indie hackers walks the full list.
Yes. Prufa runs from a URL, a CLI, an HTTP API, or as an MCP server your AI coding agent calls directly (prufa_run_audit). The free audit needs no key or card. If you build with Claude Code or Cursor, you can have the agent verify what it just built and read the report back as JSON — no test suite to author and no QA hire required.
One URL, no signup, results stream live. The same audit we ran on 49 Show HN launches — pointed at your site.