1.5M agent API keys
Moltbook, an AI-agent social network, shipped a Supabase key in its browser code with no row-level rules — exposing 1.5M agent API keys, 35K emails and private messages.4
▮ Early access · Pre-launch
Find where attackers can noclip through your app.
A plain-English security check for vibe-coded apps, AI agents and browser games built with Lovable, Bolt, Replit, v0, Cursor or Claude Code. Every finding comes with a copy-paste fix prompt for your AI builder — and a button to re-check it.
Free scan waitlist, or get a $19 report now. Founding members lock in launch pricing. We only scan apps whose owners verify them.
Level 1 · The problem
Your app looks fine. That's the trap.
AI builders ship working apps fast — but "it works" isn't "it's locked." Keys end up in browser code, databases stay readable by anyone, and agent endpoints answer strangers. These are real, public incidents:
≈60%
of 5,600+ public vibe-coded apps scanned by Escape.tech had critical flaws — plus 400+ exposed secrets.1
170 / 1,645
Lovable showcase apps had database endpoints readable without proper access rules (CVE-2025-48757).2
≈$600K
in AI credits burned by attackers via a key pulled from METR's vibe-coded agent dashboard — over three weeks.3
18,697 user records
A Lovable-hosted EdTech app had 16 vulnerabilities (6 critical) from inverted auth logic, exposing student accounts.5
"guys, i'm under attack"
A Cursor-built SaaS founder: "maxed out usage on api keys, people bypassing the subscription… I'm not technical." The app shut down days later.6
Attackers don't wait for you to launch publicly — METR's dashboard was found through new-site reconnaissance.3
Level 2 · How it works
Three steps. No security degree.
-
01
Paste your URL
Drop in your app's link. We run a quick surface check — the same things any visitor's browser can already see.
-
02
Prove it's yours
Add a small file, a meta tag, or a DNS record. We give you a one-line prompt so your AI builder adds it for you. Deeper checks unlock only after this.
-
03
Get your report + fix prompts
Every issue in plain English: what it is, why it matters, a copy-paste prompt for your builder, and a re-check this fix button so you know it actually worked.
Level 3 · What we check
Every wall attackers like to walk through.
Surface Before verification
Read-only. Only what any browser can already fetch. We show categories, never secret values.
- AI model & API keys in browser code — OpenAI, Anthropic, Gemini, Stripe live keys, Supabase service keys. We tell you which keys are meant to be public (like Supabase anon or Firebase web config) and which are dangerous.
- Public source maps that hand out your original source code.
- Exposed files like
.envand.git. - Security headers & HTTPS/TLS — the basic armor your host may not add.
- CORS — whether any website can make logged-in requests to your app.
- Email spoofing records (SPF, DKIM, DMARC) so nobody can send email as you.
Deep After you verify ownership
Still read-only and minimal: no writes, no deletes, no brute force.
- Open Supabase / Firebase data — tables, rules and storage buckets readable by anyone.
- AI agent Unprotected agent & API endpoints — routes or MCP servers that answer without a login.
- AI agent AI calls straight from the browser — the pattern that leaks model keys and burns credits.
- Game Editable leaderboards & scores — score tables that accept whatever the client sends.
- Game Exposed admin routes & game keys in your bundle.
- Rate-limit hints on login and AI endpoints, checked conservatively.
Level 4 · Sample report
What a finding looks like.
BOSS · CRITICAL
Your OpenAI key is visible in your app's browser code
What this means
Anyone who opens your game can copy this key and use your AI credits. You'd only find out when the bill arrives.
Fix prompt for Lovable (also available for Bolt, Replit, v0, Cursor, Claude Code)
Move all OpenAI API calls out of the frontend into a Supabase Edge Function. Store the key as a secret named OPENAI_API_KEY, never in client code. Have the frontend call the edge function instead, and require the user to be logged in. Then remove the old key from every file in the project.
Also: rotate the old key in your OpenAI dashboard — once a key has been public, treat it as stolen.
Demo button — shows how re-checking will work.
BOSS · HIGH
Anyone can post a score to your leaderboard — the
scores table accepts inserts from logged-out visitors. (Example)
Level 5 · Pricing
Simple pricing. Founding members lock it in.
Free scan and Watch ($9/mo) are waitlist reservations, so nothing is charged. The $19 Launch Check is available now.
Free scan
$0
- Surface check + security score
- Issue categories found
- Your top fix shown in full
Recommended
Launch Check
$19 one-time
- Full report incl. deep checks
- Fix prompt per issue for your builder
- "Re-check this fix" on every finding
- Unlimited re-scans for 30 days
Owner-verified apps only. Report within 48 hours of verification.
Watch
$9 / month
- Weekly re-scan
- Alert when a deploy brings back a leaked key or open table
- Alerts for new subdomains of your app
Founding-member offer: reserve before launch and your price stays locked for as long as you're a customer.
Level 6 · FAQ
Questions.
Can I check someone else's app?
No. Anything beyond what a normal browser already sees requires proof that you own the app (a file, meta tag or DNS record). By verifying, you confirm you own the app or are authorized to test it. If the proof is removed, deep checks stop.
Is this a penetration test?
No. NoClip Check is an automated, read-only check for common misconfigurations. It's a strong first line of defense, not a replacement for a professional security audit, and it can't promise your app has no bugs.
Do you store my secrets?
No. Any key we find is masked in the report (only the first few and last 4 characters are shown). We store the finding, not the value. When we check whether a table is readable, we record "readable: yes/no" — we don't keep your users' data.
Will the scan change or break my app?
No. Checks are read-only and minimal: we never write, delete or brute-force anything, and we keep request rates low.
My builder already has a security scan. Why this?
Use it — built-in scans are great. NoClip Check adds checks for AI-agent apps and games, works the same across builders and hosts, and closes the loop: fix prompt, then re-check that exact issue so you know the fix worked.
Isn't my Supabase or Firebase key supposed to be public?
Some are! The Supabase anon key and Firebase web config are designed to be public — your data rules are what protect you. We tell you which keys are safe to show and which (like service keys or AI model keys) are not.
What happens when I reserve?
You join the early-access list. No card, no charge. We'll email you when your spot opens, with your founding price locked.
Final level
Patch the walls before someone clips through.
Reserve early access. Founding members get launch pricing locked in.
★ Spot reserved ★
You're on the founding-member list. We'll email you when your check is ready — with your launch price locked.