VibeGuard MCP Server

📦 Install

Option 1: From npm (when published)

npx vibeguard install

Option 2: From GitHub (works now!)

git clone https://github.com/yagyeshVyas/VibeGuard.git
cd VibeGuard
npm install
npm link    # makes 'vibeguard' available globally

Then use it:

vibeguard scan          # scan your project
vibeguard dashboard     # beautiful terminal dashboard
vibeguard fix           # auto-fix 43 rule types
vibeguard pre-deploy    # 13-gate deployment check
vibeguard firewall "your prompt here"   # AI firewall

Wire into Claude Code (from GitHub clone)

# In your project directory:
echo '{ "mcpServers": { "vibeguard": { "command": "node", "args": ["/path/to/VibeGuard/bin/cli.js", "mcp"] } } }' > .mcp.json

Or use the global Claude Code command:

claude mcp add vibeguard -- node /path/to/VibeGuard/bin/cli.js mcp

Wire into Cursor / Windsurf / other MCP clients

Add to your MCP config file:

{
  "mcpServers": {
    "vibeguard": {
      "command": "node",
      "args": ["/path/to/VibeGuard/bin/cli.js", "mcp"]
    }
  }
}

Wire into Codex CLI

# Add to ~/.codex/config.toml:
[mcp_servers.vibeguard]
command = "node"
args = ["/path/to/VibeGuard/bin/cli.js", "mcp"]

vibeguard scan          # scan your project
vibeguard dashboard     # beautiful terminal dashboard
vibeguard fix           # auto-fix 43 rule types
vibeguard pre-deploy    # 13-gate deployment check

VibeGuard is a free, zero-dependency security and privacy scanner that runs entirely on your machine. It reads your code files locally, checks them against 699 security rules, and tells you exactly what's wrong and how to fix it — all without sending a single byte to any server.

It's built specifically for the problems AI coding tools create:

• Hardcoded API keys visible in browser dev tools

• Supabase/Firebase databases open to the world (no RLS)

• Admin endpoints with no authentication

• SQL injection, command injection, XSS

• Secrets committed to git history

• Hallucinated npm packages that don't exist

• User data sent to AI APIs without redaction

• Analytics tracking without consent (GDPR violation)

• WebRTC IP leaks, clipboard theft, geolocation without consent

Even if an AI is fully jailbroken, it cannot leak data. Each layer runs automatically when the MCP server starts.

If you use Supabase, Firebase, or Convex, VibeGuard checks:

• RLS missing — table created but row-level security never enabled

• RLS fake policy — USING (true) — looks enabled in dashboard, protects nothing

• PERMISSIVE policy — allows access unless another policy blocks it

• Service role key in client — bypasses RLS entirely

• Firebase rules open — .read: true or .write: true

• Firebase rules any-auth — auth != null means ANY logged-in user sees ALL data

• Convex public query — exported query with no auth check

OpenAI, Anthropic, Stripe (live + restricted), AWS, GitHub, Slack, GitLab, Twilio, SendGrid, Mailgun, Telegram, npm, Resend, GCP, Azure, Cloudflare, Datadog, Sentry, Algolia, Vercel, Netlify, Heroku, DigitalOcean, Notion, Linear, Contentful, PlanetScale, Vault, Doppler, Grafana, Jira, PagerDuty, New Relic, Splunk, Braintree, Square, Paddle, MUX, Mailchimp, Postmark, Sendinblue, Twitch, Patreon, Discord, Bitbucket, JFrog, LaunchDarkly, Rollbar, Mapbox, Twilio SID, and more.

Plus: connection strings with passwords, .env not in .gitignore, secrets in localStorage/sessionStorage, secrets in URL params, secrets in client-side config files.

• SQL injection — string concat, template literals, f-strings, format(), % formatting, raw queries across Prisma, Drizzle, Sequelize, TypeORM, Kysely, MikroORM, Turso, Hono, Go, Python, PHP, C#, Kotlin, Rust

• Command injection — exec/spawn with user input, template literals, backtick interpolation

• Code injection — eval, new Function, setTimeout/setInterval with strings, vm.runInContext

• XSS — dangerouslySetInnerHTML, v-html, innerHTML, outerHTML, document.write, insertAdjacentHTML, Angular DomSanitizer bypass, Rails html_safe, Laravel {!! !!}

• SSTI — template render with user input (EJS, Pug, Handlebars, Mustache)

• LDAP, XPath, NoSQL $where, header injection, log injection, CSV formula injection, PDF injection

• SSRF — fetch to internal IPs, cloud metadata service (169.254.169.254), DNS rebinding

• JWT alg:none, weak JWT secret, JWT decoded without verify

• Hardcoded admin credentials

• Password compared with == (plaintext)

• Session fixation (no regenerate after login)

• No rate limiting on login/register

• Password in query string

• Missing MFA enforcement

• OAuth state parameter empty/static

• Token in URL

• No logout endpoint

• Prompt injection in tool descriptions

• AI agent can execute code, deploy, install packages, access secrets, modify auth

• LLM output used as SQL, shell, DOM, file path

• Data exfiltration via agent

• Memory poisoning

• Training data leakage

• Agent loop without iteration cap

• Tool result injection

• Missing max_tokens, temperature too high

• Browser API key exposure

• Secure prompt analysis — analyzes your prompt BEFORE code generation

• Slopsquat detection — hallucinated npm packages that don't exist

• Typosquat detection

• Postinstall scripts with curl/wget/eval

• Unpinned Docker images (:latest)

• GPL/AGPL license risk

• Missing integrity hashes

• git+http dependencies

• Crypto miners in dependencies

• Missing .npmrc

• Terraform: S3 public, IAM wildcard, SG 0.0.0.0/0, RDS public, no encryption, no KMS rotation, hardcoded keys, no backup, no CloudTrail

• Kubernetes: run-as-root, no readOnly FS, hostPath, hostPID/IPC, no NetworkPolicy, privileged, SYS_ADMIN, no seccomp, no probes, no security context

• Docker: no USER directive, no HEALTHCHECK, privileged, no resource limits, secrets in env, :latest

• Docker Compose: no resource limits, secrets in environment, privileged, host network, weak restart policy

• GitHub Actions: script injection, write-all permissions, unpinned actions

• CI/CD: GitLab secrets in YAML, Jenkins credentials, CircleCI context, AWS CDK secrets, CloudFormation, Helm, Ansible, Pulumi

VibeGuard is the only scanner that protects user privacy, not just security:

• WebRTC IP leak, clipboard reads, geolocation, camera/microphone

• Canvas fingerprinting, device fingerprinting, Bluetooth scanning

• Analytics without consent (Google Analytics, Mixpanel, Segment, PostHog, Amplitude, Facebook Pixel, Hotjar, FullStory)

• Cookies without banner (ePrivacy/GDPR)

• Sentry with PII, PII in logs, health data access, biometric data

• Background location, contacts/photos/SMS/call log access

• EXIF location leak, keystroke listeners, SRI missing

vibeguard privacy-audit    # inventory all PII collection, storage, transmission
vibeguard net-audit        # map every outbound HTTP call (where does data go?)
vibeguard ai-guard         # detect user data sent to AI APIs without redaction
vibeguard privacy-policy   # auto-generate a privacy policy from code analysis
vibeguard csp-generate     # auto-generate Content-Security-Policy from code
vibeguard firewall <prompt>    # AI firewall — inspect prompt before sending to LLM
vibeguard agent-guard     # show blocked paths, commands, domains
vibeguard exfil-check <data>   # check data for PII/secrets before sending externally
vibeguard dep-firewall <pkg>    # check package name before npm install
vibeguard sandbox <code>        # execute code in zero-trust sandbox
vibeguard output-guard <text>   # sanitize AI response for secrets/PII
vibeguard env-lock        # lock environment variables — hide secrets from AI
vibeguard self-check      # verify VibeGuard hasn't been tampered with
vibeguard pre-deploy      # run all 13 security gates before deployment

PII Detection & Redaction

const { detectPII, redactText } = require('vibeguard/pii');

const text = 'Contact john@example.com or 555-123-4567. Card: 4242424242424242';
const detected = detectPII(text);  // [{ type: 'email', ... }, { type: 'credit-card', ... }]
const safe = redactText(text);      // 'Contact [REDACTED_EMAIL] or [REDACTED_PHONE]. Card: ****4242'

Detected types: email, phone, SSN, credit card (Luhn-validated), AWS keys, JWTs, private keys, IP addresses, IBANs, MAC addresses.

Run all 13 security gates before deployment. If ANY gate fails, deployment is blocked.

vibeguard pre-deploy          # run all gates
vibeguard pre-deploy --strict  # fail on warnings too
vibeguard pre-deploy --json    # JSON output for CI/CD

VibeGuard never sends your code, secrets, API keys, or personal information anywhere.

No telemetry. No analytics. No tracking. No data collection. No uploads.

vibeguard preset nextjs    # Next.js + React
vibeguard preset django    # Django + Python
vibeguard preset rails     # Ruby on Rails
vibeguard preset spring    # Spring Boot + Java
vibeguard preset aws       # AWS / Terraform
vibeguard preset gcp       # Google Cloud
vibeguard preset azure     # Azure
vibeguard preset api       # REST / GraphQL API
vibeguard preset mobile    # React Native / mobile
vibeguard preset fullstack # all rules (maximum coverage)

vibeguard init-ci          # GitHub Actions (SARIF + fail on CRITICAL)
vibeguard install-hook     # git pre-commit hook (blocks on critical)
vibeguard install-hook-post # PostToolUse hook (auto-scan after AI edits)
vibeguard pr-comment        # GitHub PR comment format (markdown)

Templates included for: GitLab CI, Jenkins, CircleCI, Azure Pipelines.

Languages (14): JavaScript, TypeScript, Python, Go, Java, Ruby, PHP, C#, Rust, Kotlin, Swift, Bash, SQL, YAML

Frameworks (20+): Next.js, React, Remix, Astro, SolidStart, SvelteKit, Nuxt, Django, Flask, Rails, Laravel, Spring Boot, NestJS, Fastify, Express, Hono, tRPC, Gin, Echo, Drizzle, Prisma, Sequelize, TypeORM, Kysely, MikroORM, Convex, Turso, Clerk, Auth.js, Supabase, Firebase, Stripe, Polar, LemonSqueezy, Resend, Upstash, Pinecone, PostHog, Uploadthing

Compliance: SOC 2, PCI-DSS, HIPAA, GDPR, ISO 27001, EU AI Act

npm install
npm test          # 236 tests
npm run lint      # 0 errors
npm run coverage  # coverage report

VibeGuard catches the mechanical security holes that AI tools leave behind. It does NOT:

• Prove your app is safe or leak-proof

• Track personal data end-to-end through your app automatically

• Judge business logic flaws

• Replace a real security review for anything touching money, auth, or personal data

It raises the floor fast — catching the holes that AI tools create by default.

MIT — Free forever. No ads. No tracking. No data collection.