In this video· click a step to jump to it
Test recording
Agent Nexus — app walkthrough
57,771 ms · 0 pixels changed · matches baseline
We built this regression test for agentnexus.app — it's yours to keep, free.
Take this test with you
This is a real Playwright test, recorded against agentnexus.app. Claim it into your own workspace and re-run it on every deploy — free.
export async function test(page, baseUrl, screenshotPath, stepLogger) {
const shot = (n, slug) => screenshotPath.replace('.png', `-${n}-${slug}.png`);
const SCROLL_THROUGH = "(async () => { const ease = t => 1 - Math.pow(1 - t, 3); const max = Math.min(4000, Math.max(0, document.body.scrollHeight - window.innerHeight)); if (max < 40) return; const anim = (a, b, ms) => new Promise(res => { const t0 = performance.now(); const step = now => { const p = Math.min(1, (now - t0) / ms); window.scrollTo(0, a + (b - a) * ease(p)); p < 1 ? requestAnimationFrame(step) : res(); }; requestAnimationFrame(step); }); await anim(0, max, 1400); await new Promise(r => setTimeout(r, 350)); await anim(max, 0, 800); })()";
await page.context().setExtraHTTPHeaders({ 'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/132.0.0.0 Safari/537.36' });
await page.route('**/cdn-cgi/scripts/**', function (r) { r.abort(); }).catch(function () {});
// ─── Scenario 1: public home ─────────────────────────────────────────────
stepLogger.log('Scenario 1: Agent Nexus home');
await page.goto(baseUrl, { waitUntil: 'domcontentloaded' });
await page.waitForLoadState('networkidle', { timeout: 2500 }).catch(function () {});
await page.locator('h1, main').first().waitFor({ state: 'visible', timeout: 8000 }).catch(function () {});Checks run
Notes from the demo run
AI-generatedWe signed into Agent Nexus with Google and walked ten authenticated surfaces: the registry, three live searches, an API contract, an MCP server contract, the submission form, API keys and pricing. The entry pages are the strongest thing here. Each one renders a machine-actionable contract in one screen — endpoint, auth mode, tags, docs link, measured latency and capabilities — which is exactly the shape an agent needs and almost never gets from a normal API landing page. The free tier is genuinely usable: signed in, we could browse and search everything without a card.
- Entry pages read like a contract, not a brochureENDPOINT, AUTH, TAGS, DOCS, LATENCY and CAPABILITIES in a single frame, with a VERIFIED date stamp. The Stripe entry showed a measured 49ms latency — that freshness claim is the differentiator and you put it on-screen.
- Search is instant and genuinely narrowingTyping 'resend', 'notion' and 'github' each returned a tight, readable result set with no reload and no spinner. This is the fastest path into the registry by a wide margin.
- The submission form asks for the right thingsAuth params as name:location pairs, input/output formats, rate limit, pricing. Most directories ask for a name and a URL; asking for the calling contract is why your entries are actually usable by an agent.
- Keys are honest about the free pathThe API keys page states the anonymous cap (100/day), the free-key cap (1,000/day) and that existing keys auto-upgrade on subscribe. No dark pattern.
- /registry renders all entries in one 25,910px pageWe measured the document height signed in: 25,910px with no pagination or virtualisation. Search rescues it, but the default landing state is a very long scroll and a lot of DOM. Paginating or virtualising the default view would make the first impression much faster.
- Near-duplicate entries for the same surfaceA search for 'github' returned both /entry/github-mcp and /entry/github-mcp-server, and both /entry/deepwiki-mcp and /entry/deepwiki-mcp-server, verified a day apart. Since freshness is the pitch, a dedupe pass on endpoint URL at review time would protect it.
- submit (final submission)·We filled the contract form but did not submit - we would not put a test entry into your moderation queue.
Share this run
Proof your app is not AI slop
This test was built for agentnexus.app — claim it free
One click copies the test and baseline screenshots into your own Lastest workspace. Re-run on every deploy and catch regressions before your users do — free, no card required.