I Built a Free Website Audit That Gives Claude-Ready Fix Prompts (Here's the Tech Stack)

A developer built a free website audit engine that generates copy-paste fix prompts for AI coding assistants like Claude, Cursor, and GPT. The tool runs inside a Next.js App Router project deployed on Vercel, processing 2,000 lines of code across six category analyzers to produce structured JSON output. A standalone Node/Express microservice on Railway handles screenshot capture, overcoming the 10-second timeout limit of Next.js serverless functions.

Every week there's a new "AI-powered website audit" tool. Most of them are Lighthouse wrappers that return the same generic advice: "Improve your page speed" and "Fix your meta tags." I wanted something different. Not advice. Code. So I built a 6-dimension audit engine that ends every finding with a copy-paste prompt for Claude, Cursor, or GPT. Run a URL through it, and 90 seconds later you get: The engine is live right now. Let me walk through the tech stack. The audit engine runs inside a Next.js App Router project deployed on Vercel. The core analysis pipeline is ~2,000 lines across 6 category analyzers: app/api/audit/route.ts → Entry point for scans lib/deep-audit-orchestrator.ts → Orchestrates 6 analyzers lib/audit-engine.ts → Main scoring engine Lighthouse, signals, crawl lib/keyword-gap-analyzer.ts → Competitor keyword comparison The engine processes everything server-side and returns structured JSON. The frontend just renders what the API gives it. The free scan returns a quick 3-dimension preview, while the deep audit runs as a POST to /api/audit/ id /deep when a user starts the trial. Screenshots were the hardest piece. Next.js serverless functions have a 10-second timeout — not enough for headless Chromium. Solution: a standalone Node/Express service deployed on Railway with Docker that auto-discovers 5 pages and captures viewport + full-page WebP screenshots. Key microservice features: The most important design decision was how to present fix prompts. Most audit tools give you a paragraph of advice. Our prompts look like this: "Fix meta descriptions and title tags on my site. URL: {YOUR SITE URL}. Generate unique 155-character meta descriptions for every page. Keep title tags under 60 characters. Output a table: URL | Current Title | New Title | Current Meta | New Meta" Copy. Paste into Cursor/Claude/GPT. Hit enter. Spend 15 minutes reviewing instead of 2 hours figuring out what to fix. The barrier isn't "knowing your site has issues." It's knowing what to type into your AI agent. Every audit tool should end with something you can paste. ✅ 6-dimension scoring composite 36/F on our test run ✅ 8+ issues with revenue impact estimated $6,500/mo loss ✅ 8 AI-ready fix prompts per audit ✅ Competitor comparison tables ✅ Card-required 3-day free trial checkout works — no caveats ✅ Free scan: no email, no account, 90 seconds Run any URL: outboundautonomy.com/ audit 3-day trial with full deep audit: outboundautonomy.com/trial Built with Next.js, Vercel, Railway, Stripe, and a distributed agent team. Questions? Drop them below.