MakerGuardHow it works

For Supabase-powered apps

Find out what strangers can access, before an attacker does.

Your app might look secure. But can someone read another user’s data, access private files, or find exposed credentials?

Your live app URL

Free · Read-only security checks · No app data stored

MakerGuard tests your live app for real-world data exposure and gives you evidence-backed findings, not a checklist or a code review.

Fixed something? Run the scan again and watch the verdict change.

What we check

Quick Scan

Runs from your link alone.

Public data exposureQuick Scan

Can strangers read data without logging in?

File exposureQuick Scan

Can someone retrieve the files your app itself uses, without logging in?

Exposed credentialsQuick Scan

Are privileged keys present in shipped frontend code?

Deep Scan

Runs with one test sign-in.

Cross-user accessDeep Scan

Can one user access another user’s records or files?

Settling open answersDeep Scan

The answers Quick Scan left ambiguous, like whether an empty answer means an empty table or a protected one.

Deep Scan tests tables and views through the data API, signed in as a test user. It does not test database functions.

Built for web apps using Supabase, including apps created with Lovable, Bolt, and other AI coding tools.

So far, MakerGuard has examined 61 real AI-built apps.

What you are probably wondering

Supabase already warns me about unprotected tables. Why do I need this?

The advisor reads your database settings. It never visits your live app, never tries a request as a stranger, and never signs in as one user to reach for another user’s data, which is where the worst leaks live. It also cannot see your shipped website code, where a master key would sit. This performs those tests, on any live app, whichever tool built it.

Why not just ask ChatGPT?

ChatGPT, Claude and Gemini can only judge what you paste to them. They cannot send the requests, sign in as a second user, or open your files as a stranger. They give opinions. This runs the tests.

Why should I trust the results?

Every Issue in your report is a request that actually came back with data, and the evidence is printed in its details. A test that could not be performed is marked Not checked, never guessed, and a name that is not a real table comes back Not checked, not an issue. No answer is given without the attempt behind it.

Who is behind MakerGuard?

Swati, a solo founder. Ten years at Microsoft, six of them in data security. She builds MakerGuard in public, and the contact link at the bottom of this page reaches her directly.

How much does it cost?

Nothing. It is free.

Do you store my app, my key, or my results?

No. The details you type here and the results you see are not sent to MakerGuard and are not stored by MakerGuard. The tests run in your browser, talking directly to your Supabase. If you paste your app link, the link goes to the MakerGuard reader: it reads your app's public code files, sends back what it finds, and stores nothing.

Is any of my stuff used to train a model?

No. Your key and your results are never sent to MakerGuard or stored by it, so there is nothing of yours to train on.

Is this AI?

No. There is no AI in the examiner. It runs deterministic checks and reports what came back, so a run today and a run after your fix can be compared like for like.

Questions, or a result you do not believe?