Website Precheck runs six checks — Security, Governance, Privacy, Speed, Accessibility, and SEO — against any URL, then writes the actual code to fix each problem on that specific site. One command: npx precheck. Runs in your terminal, as a Claude Skill, or as an MCP server.
A real run against our own homepage: it scores all six areas, then hands you the exact command to fix what it found — filename and all. Here's one of the findings.
$ ▍
We looked hard at the field — check-my.website, Auditzo, ObservePoint — before building. Here's the honest difference.
Every scanner we checked hands you "a prioritised fix-list for your developer." Website Precheck generates the actual code — the real Cache-Control block, the real compression command with the real filename it found on your site.
None of them say what device or network they tested on — "4G" alone covers several throttling presets. We state the exact device, RTT, throughput, and CPU slowdown Lighthouse actually used.
Free and open source. Enterprise platforms like ObservePoint run $599–$2,400/mo. Every network call goes through one audited function — no account, no tracking, readable before you run it.
Every check exists because it caught something real on a live, high-stakes client site under security review — a redirect-to-private-address bug, a directory-escape, a vague speed claim. We fixed it, then built the check so it can't come back.
Tools change — re-verify these specifics before repeating them. The point isn't the exact prices; it's that the free scanners stop at advice, and Website Precheck finishes the job.
It names the exact file it found, explains the problem without jargon, and attaches the fix.
Headers, cookie flags, HTTPS — what's protecting visitors and what's missing.
What cookies and data the site itself sets, and whether that's disclosed.
Which third-party trackers the page exposes visitors to.
Lighthouse performance & Core Web Vitals — against stated, exact test conditions.
An automated WCAG 2.1 A/AA pass via axe-core, with what to fix.
Finds every JSON-LD block and scores how complete it is — not just "present."
The terminal, the Claude Skill, and the MCP server run the real checks — actual Lighthouse, axe-core, the scripts. The ChatGPT and Gemini builds follow the same method and reason over your site: expert guidance, not a measured run.
✓ Runs the real checks
No account, no AI. One command and it prints what to fix.
Clone the repo, copy a skill folder into ~/.claude/skills/, then just ask Claude — it runs the real scripts and fixes what it can.
9 tools any MCP-capable agent can call — the full report plus each individual check — returning structured JSON, not text to re-parse.
Same method, as a chatbot
Open the kit, copy its Name + Instructions into ChatGPT → Explore GPTs → Create, and it coaches you through the same checks and fixes.
Open the ChatGPT kit →Open the kit, paste it into Gemini → Gems → New Gem, for the same guidance inside Google's assistant.
Open the Gemini kit →Straight talk: the chatbot builds (ChatGPT, Gemini) reason from the method and whatever browsing the assistant allows — that's expert judgment, not a measured run. For exact numbers and real fixes, use the terminal, the Claude Skill, or the MCP. Everything lives in the GitHub repo.
Clone it, read every line, and run it against your own site today. More free skills are on the way from Pacific AI Labs.
