MCP-native QA for AI-built apps

A green you can
actually trust.

Your AI says "done." Nocticas is the part that can't tell you what you want to hear: an agent drives a real browser through your app and returns an honest pass/fail with screenshots, then pins that passing run into a free, deterministic gate with no AI in the loop — so it can't drift, can't cheat, can't gaslight you. A green means green.

MCP-native · pinned gates run free · tests that can log in · weak points on every run · no false greens

objective: "log in, create a client named Acme, confirm it appears"
goto /login → fill credentials → submit
wait for dashboard, click "New client"
fill "Acme", save → assert row "Acme" visible
PASS · 3 steps · pinned → re-runs free20 credits · haiku

And it sees what a code scanner can't. Already past your login in a real browser, Nocticas flags runtime weak points — leaked secrets, missing headers, sessions that survive logout, force-browse — free on every run. See what we flag →

The obvious question

"Why not just give Claude a Playwright MCP?"

Because a Playwright MCP is a capability, not a verdict. It drives a browser — but the same agent that wrote your feature is the one grading it, live, with the model in the loop. That's exactly where an agent hard-codes the test, asserts on what it already knows passes, and hands you a green. You don't get a verifier. You get the author marking its own homework.

Nocticas runs the smart agent once, then pins that exact walkthrough into a deterministic re-play with no AI in the loop — it physically can't drift or please. Plus a real OTP/magic-link inbox so it actually gets past your login. Capability is cheap. A verdict you can trust is the product.

How it works

Author with the agent. Pin to a $0 gate. Let your agent close the loop.

Two engines, one workflow — exploratory authoring that costs a flat 20 credits, then a repeatable gate that runs free within your allowance.

01 / AGENTIC

Describe it in English

Give the agent a natural-language objective. Claude drives a real Chromium through the flow like a human QA tester — through your login, your forms, your checkout — and returns an honest pass/fail with screenshots. Declare post-conditions and they run as real assertions; one fails, the verdict flips.

02 / DETERMINISTIC

Pin it to a gate that can't gaslight you

Freeze the passing run into a model-free step script. No tokens, no drift, hard assertions — re-runs are free within a generous monthly allowance, on every plan. The same gate that guards your weekend build grows into your team's check on every push.

03 / MCP-NATIVE

Let your coding agent close the loop

One line, and Claude Code or Cursor verifies exactly what it just shipped — reads the verdict, patches its own code, and calls Nocticas again. Loop until green, no human handoff.

# add the server, then your agent does the rest
claude mcp add --transport http \
  --header "x-nocticas-key: <your-key>" \
  nocticas https://app.nocticas.com/api/mcp
Built for agentic-native builders

Everything you need to trust an AI-built app

Especially the things AI-built apps are full of — logins, flaky dependencies, UIs that drift, and exposure you can't see.

MCP-native

A first-class MCP server, one-line install, an "Add to Cursor" deeplink. The agent that built your feature calls Nocticas to verify it — no human in the loop.

No false greens

Pinned re-runs are exact-replay with no AI in the loop — they can't drift or tell you what you want to hear. Mocked, replayed, or self-healed runs are loudly stamped, never silent. (Self-healing rebinds only on a unique, stable id and flags the run — healing that can lie is just a slower false green.)

Tests that can log in

You can't test the functionality without getting past the login screen — so we got past it. A built-in test inbox catches real OTPs and magic-links, so auth-gated flows actually get exercised. Zero config.

Two engines, one flow

Agentic to explore, a model-free deterministic gate to ship. The agentic verdict is a flat 20 credits; the pinned gate re-runs free within your allowance — the right cost at each stage.

Weak points, flagged free

Already inside a real browser and past the login, Nocticas surfaces the runtime exposure a code scanner can't see from outside: missing security headers, insecure cookies, exposed secrets (redacted), mixed content, source maps, verbose error leaks. The runtime complement to a static scanner — a second pair of eyes, not a security certificate. Findings never flip your verdict. Free on every run; paid plans add deep exposed-secret scanning on domains you verify, plus scheduled monitoring that pings you the moment a new weakness appears.

Transparent cost

Every run shows its exact credit cost up front. Prepaid credits and hard caps — your agent physically can't run up a surprise bill.

Ship with a net

Pin it once. Gate every push.

Drop the gate into CI and Nocticas re-runs your pinned flows on every pull request — catching regressions in the paths that matter before they merge. Deterministic, no AI in the loop, and free to run on public repos.

# .github/workflows/nocticas.yml
- uses: Low-Latency-Labs/nocticas-action@v1
  with:
    api-key: ${{ secrets.NOCTICAS_API_KEY }}
    gate: .nocticas/checkout-flow.json
    target: ${{ vars.PREVIEW_URL }}

A red verdict fails the build. Works with any CI via a one-line script — no GitHub Action required.

We bet our own releases on it. We don't ship a green we don't trust — so we built the tool that proves it.

Pricing

Pinned gates run free within your allowance. Agentic runs are a flat 20 credits.

Prepaid credits, hard caps, no surprise bills. Usage-based, not seat-based — because an AI agent is often the one using it.

Free
$0
  • 500 free deterministic runs / mo
  • 150 trial credits for agentic runs
  • Core engines + MCP + test inbox
  • No card required to start
Start free
Starter
$19 / mo
  • 10,000 free deterministic runs / mo
  • 500 credits / mo (≈25 agentic runs)
  • 1 GB artifact storage
  • Credit packs that never expire
Get started

Also: Team $99/mo (2,500 cr · 50k det runs · 10 GB) · Scale $249/mo (7,500 cr · 200k det runs · 100 GB) · Enterprise. Live details on the pricing page.

Stop getting gaslit by your own agent.

Get a green you can actually trust — and a free map of where your app is exposed — on the very run you were already going to do.

Start free →