Malinois Scan

Passive security check for an app you just shipped. One command, no signup, no account, no agent installed.

npx malinois-scan https://my-app.example --i-own-this

  Grade D  50/100   my-app.example

  3 issues found

  CRITICAL  A database table is open to the public
           What it means: Anyone on the internet can read this table directly — no login
           required. If it holds names, emails, or other customer info, that data is
           exposed right now.
           What to do:    In Supabase, open your table → "Row Level Security" and turn RLS
           ON (there's a one-click button), then add a policy so only the right users can
           read it.

  CRITICAL  Your secret settings file (.env) can be downloaded
           ...

It checks what any visitor can already see from outside your app — no exploitation, no injection, no load testing, no credentials needed.

Why this exists

AI builders (Lovable, Replit, Bolt, v0, Base44) made shipping trivial. They did not make day 2 trivial. The most common ways these apps leak are boring and repeatable:

• a Supabase/Firebase table left readable by anyone
• a service_role key or API key sitting in the JavaScript bundle
• .env or .git served publicly
• source maps shipping your original code

Platform scans run when you hit publish. The leak usually arrives on the redeploy after that — which is exactly what a check in CI catches.

What it checks

Every finding comes with what it means and what to do, in plain language — not just a rule name.

Fixing what it finds

Longer write-ups for the findings people hit most, including the exact errors you'll see while fixing them:

• Supabase RLS — new row violates row-level security policy, permission denied for table, and the silent case where RLS is simply off
• Which keys are safe in a browser bundle — anon vs service_role, why NEXT_PUBLIC_/VITE_ secrets aren't secret
• Firebase security rules — Missing or insufficient permissions, and open test-mode rules that never error
• Exposed .env, .git, source maps — what to rotate after

Use it in CI (the part that actually matters)

Add this and your build fails the moment a deploy introduces a leak:

# .github/workflows/security.yml
name: Security
on:
  push:
    branches: [main]
  workflow_dispatch:

jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: parqoo/malinois-scan@v0
        with:
          url: https://my-app.example
          fail-on: high

The job summary shows the grade and every finding. Outputs grade, score, and findings are available to later steps.

Use it from your AI builder (MCP)

Don't want a terminal? Add the hosted MCP server to the assistant you already build with, then ask it: "Check https://my-app.lovable.app for leaks — I own it."

https://malinois.app/mcp

• Lovable — Chat connectors → Custom MCP → paste the URL (no authentication).

• Claude Code — claude mcp add --transport http malinois https://malinois.app/mcp

• Cursor / VS Code — one click:

  

  or add { "mcpServers": { "malinois": { "url": "https://malinois.app/mcp" } } } to mcp.json.

• Claude, ChatGPT and other chat apps — add a custom connector with the URL.

Same engine and the same boundaries as this CLI: passive only, ownership confirmation required, private and metadata addresses refused, secrets masked.

CLI options

npx malinois-scan <url> --i-own-this [options]

  --i-own-this        Required. Confirms you own the app or have permission.
  --json              Machine-readable output (for CI).
  --fail-on <level>   Exit 1 at/above this level. critical|high|medium|low (default: high)
  --no-color          Disable ANSI colour.

Exit codes: 0 clean (below threshold) · 1 findings at/above threshold · 2 scan error.

Use it as a library

import { runScan } from 'malinois-scan';
import { explain } from 'malinois-scan/explain';

const result = await runScan('https://my-app.example');
console.log(result.grade, result.score);

for (const f of result.findings) {
  const help = explain(f.rule_id, 'en'); // also: 'ko'
  console.log(f.severity, help?.title, '→', help?.fix);
}

Boundaries — on purpose

This is a defensive tool and it is built to stay one.

• Passive only. Plain GET requests any visitor could make. No exploitation, no injection, no brute force, no load testing.
• Ownership required. --i-own-this is not decoration — this is for apps you control. It is not a tool for probing strangers.
• Private ranges refused. Loopback, private, link-local and cloud-metadata addresses (e.g. 169.254.169.254) are rejected, so this can't be turned into an SSRF proxy.
• Never prints a raw secret. Findings show a masked form (sk_live_…ab12) and where it was found.
• A clean grade is not a clean bill of health. For a browser-rendered app whose backend stays invisible to a passive check, the result is marked limited and says so.

A one-time check is not monitoring

This CLI answers "is it leaking right now?". It does not answer "did something break at 3am on Tuesday?".

Malinois is the hosted side: it re-checks your app on a schedule, keeps the history, and emails you the moment a new leak appears that wasn't there before — including a plain-language report you can hand to someone non-technical. Free passive check in the browser, no signup: https://malinois.app/?src=gh_readme

FAQ

Does it need access to my code or accounts? No. It only reads what your deployment serves publicly.

Will it break anything? No. It makes a bounded number of ordinary GET requests, the same ones a browser makes when loading your page.

Can I scan my staging URL? Yes — anything you own. Prefer staging if you're unsure.

Why is my SPA graded A with no findings? If your app renders in the browser and never exposes its backend to a passive request, there may be nothing observable from outside. The output marks this as a limited scan rather than claiming you're safe.

License

MIT — see LICENSE.