Skip to content
ScanAnt

Cursor app security: check the code your AI wrote

Cursor is an editor, not a host, so an app built with it can end up anywhere and be written in any language. What stays the same is that generated code compiles and looks fine while skipping the boring parts: ownership checks, input handling, secret storage. ScanAnt scans the deployed site and, if you upload a ZIP, the source itself.

Free, no signup. We only read what your site already shows every visitor.

What we check

Each item is tagged with the scan that reports it. The quick scan is free and needs no signup. The others run only on sites you have proved you own.

  • The deployed site

    Quick scan

    Headers, HTTPS, cookie flags, public source maps and keys that shipped in the bundle, wherever you host.

  • Front-end code patterns

    Quick scan

    Likely DOM XSS, redirects driven by the URL, tokens in localStorage, login data printed to the console and unfiltered reads of private-looking tables.

  • Hard-coded secrets

    Code scan

    Keys written into source, committed .env files and secrets that fall back to a default value in code.

  • Routes without authentication

    Code scan

    Server endpoints that do powerful things without checking the caller, and records that can be opened by id without an owner check.

  • Injection patterns

    Code scan

    SQL built from pasted text, HTML inserted without cleaning, request input reaching a shell command or file path and URLs fetched without a check.

  • Vulnerable dependencies

    Code scan

    Packages in your lockfiles with published vulnerabilities, with the affected version.

  • Optional AI review

    Code scan

    A separate review of logic the pattern checks cannot judge, labelled so you know it is a possibility, not a proof.

  • Exposed files on the live host

    Verified scan

    Public .env files, .git folders, backups and debug paths, after ownership proof.

  • Quick scan: No signup. Reads only what every visitor receives.
  • Code scan: You upload your project as a ZIP.
  • Verified scan: After you prove you own the site. Read-only.

No scan can promise to find every problem. A clean result means no known problems were found on what the scan could see.

Common mistakes

  • Accepting a generated endpoint without an auth check

    Models often write the happy path first, so any caller can reach it.

  • Letting the editor create a .env and committing it

    The first push stores the secret in history permanently.

  • Taking the price or role from the request body

    The client can send any value it likes.

  • Adding a package the model suggested without checking it

    It may be outdated or carry known vulnerabilities.

  • Trusting that tests passing means safe

    Functional tests rarely try another user's record id or a hostile input.

How to fix it

  1. Add authentication and an owner check to each route that returns or changes user data.

  2. Move secrets to environment variables, rotate any that were committed and add .env to .gitignore.

  3. Use parameterised queries, escape output and validate URLs and file paths from the request.

  4. Update flagged dependencies, then rerun the scan.

  5. Paste each finding's fix prompt into Cursor so the change stays narrow.

Questions

Which languages does the code scan read?

JavaScript and TypeScript, Python, PHP, Ruby, Go and Java, plus SQL and Firebase rules. The report lists files it could not analyse instead of hiding them.

Is my uploaded code stored?

Findings store a masked value and the location of secrets, never the full secret. Check the Privacy page for what is kept and for how long.

Does my code go to an AI model?

Detected secrets are never sent to an AI model. The AI review is optional and separate.

Do I need a deployed site?

No. You can upload the project alone. A live URL adds the checks that only work on a running server.

Run the scan now

Free for any public site, no signup. Need the verified, deep or code scan? Choose the level on the scan page.

Choose a scan level →