Now in public beta

Your AI writes code. Veilguard makes sure it doesn't get you hacked.

Free security scanner for AI-generated code. Catches leaked API keys, SQL injection, broken database policies, and supply chain attacks — while you vibe.

Install FreeSee What It Catches
Works in Cursor · Claude Code · VS Code · Windsurf · Antigravity
my-app—AI Agent
Add Stripe payments to my checkout page
AI Agent

✓Done. Created app/api/checkout/route.ts with your payment logic.

scan this for security issues
Veilguard
Veilguard·gentle nudge
CriticalYour Stripe live key is sitting in your code.
sk_live_51Mrt8K2eZvKYmT...Xk9

Anyone who can see your GitHub repo — or just right-clicks “View Source” on your live site — can grab this key and charge real money on your customers' cards.

Fix→move it to your.envfile and never touch it again
fix that
AI Agent

✓Moved to .env. Gone from your code and your git history.

Veilguard
Veilguard·all clear ✓
Ask the agent…
THE PROBLEM

AI-generated code has a security problem

45%
of AI-generated code contains security vulnerabilities
Veracode, 2025
28.6M
hardcoded secrets pushed to public GitHub repos in 2025 — a 34% YoY increase
GitGuardian
1.5M
API keys leaked in the Moltbook breach — a fully vibe-coded app with zero security review
Moltbook, 2026
74
CVEs directly attributed to AI coding tools
Georgia Tech Vibe Security Radar

Vibe coding ships products in days. It also ships vulnerabilities. Tools like Cursor, Claude Code, and Windsurf generate functional code fast — but they don't check for hardcoded secrets, SQL injection, broken database policies, or malicious dependencies. Veilguard does.

HOW IT WORKS

Install once. Code normally. Sleep well.

STEP 1

Install in 30 seconds

Add one MCP config block to your IDE — the same JSON everywhere, only the file path changes. Claude Code installs with a single command.

claude mcp add veilguard -- npx -y --package=veilguard veilguard-mcp
STEP 2

Code like you always do

Your AI agent calls Veilguard's tools as it writes code — automatically once you add the rules file, or any time you ask. If everything is clean, total silence. You never know it's there.

SILENT · IN YOUR IDE
STEP 3

Get a nudge when something's wrong

Found a leaked Stripe key? Your AI agent flags it right in the chat — in plain English, with the real-world consequence — then offers you the exact fix with Pro. SQL injection in an API route? Same thing. Clean code stays silent.

PLAIN ENGLISH · FIX WITH PRO
SECRET SCANNING

60+ patterns. Zero leaks.

Catches 60+ API key patterns: Stripe, OpenAI, Supabase, Paystack, Flutterwave, M-Pesa, AWS, and more. Detects the #1 vibe coding mistake: AI adding live keys as fallback values.

Every file modification is checked in milliseconds. Before the commit even happens.

[CRITICAL]Stripe live key detected in src/lib/payments.ts:14
Found: sk_live_51M******************
[WARNING]Supabase service role key in .env.local does not match standard ignore patterns.
// Before: AI hallucinated policy
CREATE POLICY "Users can update" ON users
FOR UPDATE USING (auth.uid() IS NOT NULL);
// After: Veilguard fix applied
CREATE POLICY "Users can update" ON users
FOR UPDATE USING (auth.uid() = id);
SUPABASE SECURITY

Catches the flaws behind real breaches.

Deep analysis of Row Level Security policies. Catches the exact patterns behind the Moltbook breach (1.5M leaked keys) and Lovable CVE-2025-48757 (170 apps exposed). Detects USING(true), auth.uid() IS NOT NULL bypass, and missing policies.

Breach Context

This exact flaw caused the Moltbook breach in January 2026, leaking 1.5M API keys and 35,000 emails.

WEBHOOK SECURITY

Unverified webhooks = free money for attackers.

Finds webhook endpoints missing signature verification for Stripe (constructEvent), Paystack (HMAC), M-Pesa (IP check), GitHub, and Flutterwave. AI will happily skip these critical checks.

[CRITICAL]Stripe webhook in app/api/webhooks/route.ts missing constructEvent verification.
[WARNING]M-Pesa callback handler lacks IP allowlist validation.
WHAT VEILGUARD CATCHES

14 scanners. Every vulnerability AI introduces.

Secret Detection

Catches 60+ API key patterns: Stripe, OpenAI, Supabase, Paystack, Flutterwave, M-Pesa, AWS, and more. Detects fallback trap keys.

SQL Injection

Detects template literal injection, unsanitized req.body, command injection, NoSQL injection, IDOR, and mass assignment.

Webhook Verification

Finds webhook endpoints missing signature verification for Stripe, Paystack, M-Pesa, GitHub, and Flutterwave.

Environment Security

Checks .env is in .gitignore, detects secrets exposed via NEXT_PUBLIC_ or VITE_ prefixes.

CORS Misconfiguration

Catches cors({ origin: '*' }) on apps with authentication.

Supply Chain

Detects malicious and typosquatted npm packages against a known-bad database.

Dependency CVEs

Checks every npm dependency against Google OSV.dev for known vulnerabilities.

Auth Configuration

Validates Clerk, NextAuth, and Supabase Auth. Catches getSession() spoofing and missing rate limiting.

Security Headers

Checks CSP, HSTS, X-Frame-Options on deployed URLs.

Git Security

Scans for secrets in git history, .gitignore gaps, and tracked .env files.

Supabase RLS Audit

Deep analysis of Row Level Security policies. Catches USING(true), auth.uid() IS NOT NULL bypass, and missing policies.

Firebase Rules Audit

Analyzes Firebase security rules for allow if true, client-controlled userId, and auth-only policies without ownership checks.

App Security

Rate limiting, IDOR, password storage, file uploads, error exposure, sensitive logging, open redirects, mass assignment.

AI Rules File Scanner

Scans .cursorrules, .windsurfrules, and CLAUDE.md for hidden Unicode backdoors, base64 payloads, and malicious instructions that hijack your AI agent.

Full Security Audit

Pro

Runs all scanners, scores your project 0-100, assigns a grade A+ to F, and generates an AI-ready fix prompt. Pro only — unlimited.

REAL BREACHES

Based on real breaches. Not theoretical risks.

Moltbook2026
1.5M API keys + 35K emails leaked
Veilguard catches this with: check_supabase_rls
Lovable2025
CVE-2025-48757 — 170 apps exposed
Veilguard catches this with: check_supabase_rls
GitGuardian 20262025
28.65M hardcoded secrets in public repos
Veilguard catches this with: scan_secrets
INSTALLATION

One command. Every IDE.

Path: .claude/mcp.json (also: claude mcp add veilguard -- npx -y --package=veilguard veilguard-mcp)
{
  "mcpServers": {
    "veilguard": {
      "command": "npx",
      "args": ["-y", "--package=veilguard", "veilguard-mcp"],
      "env": { "VEILGUARD_KEY": "your_key_here" }
    }
  }
}
One command in Claude Code:claude mcp add veilguard -- npx -y --package=veilguard veilguard-mcp
Free: leave VEILGUARD_KEY empty. Pro: paste key from email.
PRICING

Free is powerful. Pro is complete.

Free

$0/mo
  • All 14 scanners (including RLS & Firebase)
  • Every vulnerability flagged — plain-English alerts
  • MCP server — works in Cursor, Claude Code, Windsurf, VS Code & Antigravity
  • Fixes & solutions — upgrade to unlock
  • Full security audit & letter grade (Pro only)
Get started
RECOMMENDED

Pro

$19/mo
  • Everything in Free
  • The exact fix for every vulnerability
  • Full audit with letter grade (A+ to F)
  • AI-ready fix prompt (paste to fix everything at once)
  • Full security audit — unlimited
  • Breach context on every finding
Go Pro
IDE SUPPORT

Works in every AI coding IDE

Cursor

Add the MCP config, ask it to scan — it just works.

Claude Code

One command to connect. Then just ask Claude to check your code.

Windsurf

Drop in the config file, restart, and you're good.

VS Code

MCP server connects via Copilot Chat in one config file.

Antigravity

Add via the MCP Settings Panel, no extra setup.

PRIVACY

100% local. Your code never leaves your machine.

Veilguard runs entirely on your laptop. It reads your files locally, matches patterns locally, and returns results locally. The only outbound calls are to Google's OSV.dev API (sends package names only, never code) and veilguard.dev for Pro license validation. No telemetry. No data collection. Open source on GitHub.

View on GitHub ↑
FAQ

Questions vibe coders ask before installing

Everything you need to know about securing AI-generated code, vibe coding security risks, and how Veilguard works.

What is vibe coding?
Vibe coding is a software development approach where developers describe what they want in plain English and an AI agent — such as Cursor, Claude Code, or Windsurf — writes the entire application. It lets anyone build production-ready apps in hours without writing most of the code manually. Over 92% of developers now use AI coding assistants, and 41% of all code written globally is AI-generated.
Is vibe coding safe?
Vibe coding produces working code fast, but 45% of AI-generated code contains security vulnerabilities according to Veracode (2025). AI tools do not check for hardcoded secrets, SQL injection, or broken database policies. Vibe-coded apps need a dedicated security scanner to catch what the AI misses before those vulnerabilities reach production.
What security vulnerabilities does AI-generated code typically introduce?
The most common vulnerabilities are: hardcoded API keys embedded as fallback values (process.env.KEY || 'sk_live_...'), SQL injection via template literals, missing Supabase Row Level Security, unverified webhook handlers that accept fake payment events, wildcard CORS on authenticated apps, and typosquatted npm packages. AI tools prioritize making code work quickly over making it secure.
How do I secure a vibe-coded app?
Scan every file for hardcoded secrets before git commits, audit your Supabase RLS policies to confirm row-level access is enforced per user, verify all payment webhooks validate signatures cryptographically, check npm dependencies for known CVEs, and review your CORS configuration. Veilguard automates all of these checks through your AI coding IDE using 14 security scanners.
What is an MCP security server?
An MCP (Model Context Protocol) security server is a tool that integrates directly into AI coding IDEs like Cursor, Claude Code, and Windsurf. It exposes security scanning functions the AI agent can call while writing code. Veilguard's MCP server adds 14 security scanners — covering secrets, SQL injection, Supabase RLS, webhook verification, supply chain attacks, and more — to any supported AI coding IDE.
Can using Cursor or Claude Code result in my app being hacked?
AI coding tools like Cursor, Claude Code, and Windsurf routinely generate insecure patterns without warning. The Moltbook breach (January 2026) leaked 1.5 million API keys from a fully AI-coded app. Lovable's CVE-2025-48757 exposed 170 apps through AI-generated RLS logic errors. AI writes fast code — without a security layer, that speed ships vulnerabilities directly to production.
Is Veilguard free?
Yes. Veilguard's free tier runs all 14 security scanners and the MCP server in Cursor, Claude Code, Windsurf, VS Code, and Antigravity, and alerts you to every vulnerability it finds. Free tells you exactly what's wrong; the fixes, breach context, and the full security audit (letter grade A+ to F plus an AI-ready fix prompt you can paste to fix everything at once) are Veilguard Pro — $19/month or $149/year.
Does Veilguard send my code to the cloud?
No. Veilguard runs entirely on your local machine. It reads files locally, matches patterns locally, and returns results locally. The only external calls are package names (never source code) to Google's OSV.dev for CVE lookups, and your license key to veilguard.dev for Pro validation once every 24 hours.

Ship fast. Ship secure.

Install Veilguard in 30 seconds. Free forever for indie developers.

claudemcp add veilguard -- npx -y --package=veilguard veilguard-mcp
Install FreeGo Pro — $19/mo