Stop asking your friends
to test your app.
Val is the QA agent you talk to in Cursor, Claude Code, or Codex. Ask Claude to check your build, Val opens a real browser, walks every flow, finds the bugs, your AI fixes them. Ship to green without bothering anyone.
Every bug a user would hit.
Broken links & 404s
Every link followed. Val flags the ones that dead-end on a 404 or a blank page before a visitor ever hits them.
Console errors
Uncaught exceptions, failed fetches, red console spew. Val watches the console on every page and reports it.
Dead buttons & forms
The button that does nothing, the form that never submits. Val clicks everything and tells you what did not respond.
Hidden & clipped text
Text trapped behind a container, cut off at the edge, invisible on its own background. Val reads what a human cannot.
Broken layouts
Overflowing containers, elements stacked wrong, a mobile view that falls apart. Val checks how it actually renders.
Broken images
The missing logo, the 404'd screenshot, the icon that never loaded. Val catches every asset that failed to show up.
Dead-end flows
Sign up, then nowhere. Checkout that loops. Val walks your key journeys end to end and finds where they break.
Janky first paint
Pages that flash, shift, or hang before they settle. Val notices the rough first impression users feel but cannot name.
Regressions
Fixed one thing, broke another? After every fix Val re-runs the whole sweep so a patch never quietly takes something down.
Detect. Fix. Verify.
Detect
Val drives a real browser through your app, on localhost or live, and clicks every link, flow and screen to find what's broken.
Fix
Each finding goes to your coding agent as a precise instruction. It edits the code in your own editor, on your own key.
Verify
Val re-runs the check, then the whole critical path, to confirm the fix held and nothing else broke. If it did, it loops again.
Detect, fix, verify, repeat, until everything is green. You just say yes.
Lives in your coding agent.
Val is an MCP server and a standalone CLI. Drop it into the assistant you already use and your AI drives it. Or skip the agent and run it straight from your terminal. Same engine, same license key, two doors.
{
"mcpServers": {
"val": {
"command": "npx",
"args": ["-y", "@nyx-intelligence/val-mcp"],
"env": { "VAL_LICENSE_KEY": "vk_your_key" }
}
}
}# install once npm install -g @nyx-intelligence/val-mcp # set your license key export VAL_LICENSE_KEY=vk_your_key # scan any URL val-mcp scan https://my-app.com # click every button on a page val-mcp exercise https://my-app.com
One agent. Five bucks.
Bring your own model key. We charge for the agent and the check library, never your tokens.
- The full check library, updated weekly
- The self-fixing detect, fix, verify loop
- Unlimited runs, on your own key
- Localhost and live URLs
- Every supported coding agent
What Val does. What it doesn't.
Honest scope. Subscribe knowing exactly what you're getting.
✓ What works
Apps behind a login?+
val_login(url, email, password) finds the email and password fields, signs in, persists the session. Every subsequent Val call runs authenticated. Use a dedicated test account, not your prod credentials.Paid funnels (signup, checkout, subscription)?+
Full Lighthouse audit?+
val_lighthouse runs the complete Lighthouse 13 audit (performance, accessibility, best-practices, SEO) and returns the category scores plus the core Web Vitals (LCP, FCP, TBT, CLS, Speed Index, TTI) and the eight worst failing audits. Desktop or mobile form factor.