npx vibecheck-launch-mcp

VibeCheck inside
your AI assistant

Run pre-launch audits without leaving Claude, Cursor, or Claude Code. One config block. Five tools. No browser tab required.

Four ways to use VibeCheck

Pick whichever fits your workflow. All four run the same checks and return the same results.

New checks deploy to the worker and appear in all four automatically — no updates needed.

What it looks like in Claude

Ask Claude what tools are available — or just tell it to run an audit. It picks the right tool automatically.

Claude — claude.ai

What vibecheck tools do you have available?

C
Thinking about available vibe-checking capabilities and tools ›
C

Five VibeCheck tools, all connected and ready to use:

vibecheck_auditFull pre-launch audit against a URL. Checks auth, payments, AI safety, security, SEO, mobile, performance, and more. Returns a Vibe Score plus a prioritized issue list. Takes 3-6 minutes.
vibecheck_quick_checkSingle-category fast audit (~90 seconds). Good when you want a targeted check — just security, just SEO, just payments — without running the full suite.
vibecheck_fix_promptTakes a completed audit ID (or specific check IDs) and generates a structured fix prompt ready to paste into Claude Code or any AI coding assistant.
vibecheck_explainExplain any specific check by its ID (e.g. auth-018, ai-002). Returns what it tests, why it matters, failure rate data, and how to fix it.
vibecheck_leaderboardShows the most commonly failing pre-launch checks across all audits, ranked by failure rate. Useful for proactively hardening before you run a full audit.

You can ask me to run any of these — or just say "audit my staging URL" and I'll handle it.

Write a message...
Sonnet 4.6

Five tools

Ask Claude to call these by name, or they trigger automatically when Claude decides they are relevant.

vibecheck_audit3–6 min

Full pre-launch audit against a URL. Checks auth, payments, AI safety, security, SEO, mobile, performance, and more. Returns a Vibe Score plus a prioritized issue list. You can scope it to specific categories or pass test credentials for authenticated flows.

vibecheck_audit({ url: "https://staging.yourapp.com" })
vibecheck_quick_check~90 sec

Single-category fast audit. Good when you want a targeted check — just security, just SEO, just payments — without running the full suite. Categories: auth, payments, database, api, frontend, mobile, performance, security, ai, seo, legal, content, production, observability, launch.

vibecheck_quick_check({ url: "https://staging.yourapp.com", category: "security" })
vibecheck_fix_promptinstant

Takes a completed audit ID (or specific check IDs) and generates a structured fix prompt ready to paste into Claude Code or any AI coding assistant. Pass a stack_hint to make the fixes specific to your stack.

vibecheck_fix_prompt({ audit_id: "clx...", stack_hint: "Next.js, Supabase, Vercel" })
vibecheck_explaininstant

Explain any specific check by its ID. Returns what it tests, why it matters, real-world failure rate data, and exactly how to fix it. Stack-aware if you pass context.

vibecheck_explain({ check_id: "auth-018", context: "Next.js App Router" })
vibecheck_leaderboardinstant

Shows the most commonly failing pre-launch checks across all VibeCheck audits, ranked by failure rate. Useful for proactively hardening your app before you even run a full audit.

vibecheck_leaderboard({ limit: 10, category: "security" })

Setup

Works with any MCP-compatible AI assistant. Requires Node.js 18+.

1

Open your MCP config file

Claude Desktop~/Library/Application Support/Claude/claude_desktop_config.json
Cursor.cursor/mcp.json (project) or ~/.cursor/mcp.json (global)
Claude Code.claude/mcp.json in your project root

Create the file if it does not exist.

2

Add the vibecheck server

Paste this into your config. If you already have other MCP servers, add the vibecheck key inside your existing mcpServers object.

claude_desktop_config.jsoncopy and paste
{
  "mcpServers": {
    "vibecheck": {
      "command": "npx",
      "args": ["vibecheck-launch-mcp"]
    }
  }
}
3

Restart your AI assistant

Fully quit and reopen it — not just close the window. MCP servers are loaded on startup.

4

Run your first audit

Type this in your AI assistant chat:

Run a VibeCheck audit on https://your-staging-url.com

The URL must be publicly reachable. No localhost.

What you get back

Claude reads the audit result and surfaces the most important issues — ready to fix in the same conversation.

Claude Desktop

VIBECHECK AUDIT COMPLETE

URL: https://your-staging-url.com

Vibe Score: 61/100 — Fix before launch

Checks run: 168 | Passed: 141 | Failed: 12 | Warned: 9 | Skipped: 6

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

CRITICAL FAILURES (2)

[sec-004] Sensitive env vars in client bundle

Found: sk_live_ pattern in /_next/static/chunks/main.js

Fix: Move secret API calls to server-side routes only

[auth-018] Admin routes not protected server-side

Found: /admin returned HTTP 200 with no auth redirect

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ Use vibecheck_fix_prompt to generate a fix for all issues

Why run audits from inside your AI assistant?

When you build with Claude Code or Cursor, your whole context is already there — the codebase, what you just changed, what you are about to ship. Running an audit from a browser tab means context-switching: you open the tab, paste the URL, wait, read the results, switch back, and try to apply the fixes while the report fades from memory.

The MCP integration keeps everything in one place. Run the audit, get the results, ask Claude to generate the fixes, and apply them — without leaving the conversation. The vibecheck_fix_prompt tool takes an audit ID and generates a structured prompt Claude can act on immediately. The gap between "found a problem" and "fixed a problem" is one tool call.

Pre-launch auditing, inside your AI assistant.

No browser tab. No context switching. Just paste the config and restart.