# ScreensDesign — studied reference

- **Source:** [screensdesign.com](https://screensdesign.com/) · studied live 08-08-2026 (headless browser + curl, no login).
- **Screen types seen:** onboarding/welcome · paywall/pricing · social proof · list/feed · detail view · search/filter (the fifteen-type list is in [references.md](references.md)).
- **Platforms covered:** iOS only. The page's own title says it: "ScreensDesign: App Design Library & AI Screen Generator — iOS Examples."

## What it is

**A paid teardown library.** ScreensDesign holds 2,622 top-grossing iOS apps (count as read live, 08-08-2026), each one broken into its onboarding, paywall and App Store screens. Every app gets one page: a single scrubbable video of the real flow, cut into named chapters with timestamps — for one app studied, "Onboarding and Subscribing 00:02–00:58", "Generating and Managing Avatars 00:46–01:59", "Exploring and Creating New Packs 02:00–03:04" — a strip of timestamped screen thumbnails under the video, the app's revenue, installs, release date and developer, a one-line description, and a grid of similar apps sorted the same way. The library also carries a natural-language search bar over the whole set ("Ask the library" — sample query on the homepage: "Show me meditation paywalls with a free trial, sorted by revenue…") and a four-group filter panel: **App basics** (category, release/updates), **Business** (installs, revenue), **Onboarding** (steps, features, permission screens, social proof), **Paywall** (type, features).

**A second product, since a "V2" launch.** Alongside the library sits **Create**: type one sentence describing an app and it generates an onboarding flow, paywall and screen set, trained on the same 2,600-app corpus, meant to be handed to an AI coding agent to build. This is new since the one-line entry already in [references.md](references.md) was written — ScreensDesign is no longer only a gallery.

**Who it is for.** The whole front page is an indie-app and growth-marketer pitch: a wall of X testimonials, a logo row of paying customers (RevenueCat, Superwall, Noom, Headway, Suno, Polymarket among them), and copy like "Every screen. Every paywall. Every dollar." It is built to answer one question — how to turn a stranger into a paying subscriber, fast.

## Free versus paid

**Free, no account needed.** Every app's card: name, developer, revenue, installs, release date, category. The filter panel itself is free to browse and use. Each app's page shows its first few screen thumbnails and the first seconds of its teardown video for free — enough to judge whether the app is worth studying.

**Paid — one plan seen, "Full Pro."** As read on the pricing page 08-08-2026: **$24.92/month billed annually at $299/year**, with weekly and monthly options too. It unlocks: the rest of every screen strip (every thumbnail past the free first few carries a plain "Unlock Pro" tag, timestamp still visible), the full similar-apps grid, "Ask the library" (the natural-language search), 2,400 Create credits a year, and MCP access. The page's own copy reads "EVERYTHING IN PRO, PLUS" above the feature list, which suggests a lower-priced "Pro" tier exists or once existed; only the one "Full Pro" card rendered when checked, so that is what this note can confirm.

**The honest read.** This is a freemium teardown library, priced like a professional tool, not a casual one. Nothing is free to export or download; everything free is free to *look at only*.

## The installable finding — confirmed and recorded, so it need not be re-checked

Frank's manual check found that `screensdesign.com/mcp`, `/api` and a nonsense path all return the identical page, byte for byte. **That is correct for a plain fetch, and it is a stale way to answer the question "is anything installable."**

**What a bare `curl` sees** (no JavaScript, confirmed 08-08-2026): `/`, `/mcp`, `/api` and a made-up path return three different byte counts overall, but `/mcp`, `/api` and the nonsense path all come back as the exact same 8,502-byte shell, same SHA-256 hash. That shell is what ScreensDesign's server hands back for any route it does not resolve server-side — the site is a Next.js app whose real routing happens in the browser, after its JavaScript runs. Checked with a plain fetch, `/mcp` and `/api` and a typo look identical, because none of them have been *decided yet*.

**What a JavaScript-executing browser sees, once it runs:** `/api` still renders nothing but the homepage — it is not a real page, just an unrouted path with nowhere to fall through to but the front door. **`/mcp` is different.** It renders 1,258 characters of real, specific content: a page titled "ScreensDesign MCP for any AI," with setup instructions for Claude Code, Claude Web, ChatGPT, Codex, Cursor, OpenClaw and others, naming a real hosted server — `https://api.screensdesign.com/v1/mcp` (Streamable HTTP transport) — behind OAuth sign-in, plus a public companion "agent skill": `npx -y skills add https://github.com/screensdesign-com/screensdesign-agent-skill`.

**Both pieces check out independently.** A direct POST to `https://api.screensdesign.com/v1/mcp` returns HTTP 401 with `WWW-Authenticate: Bearer realm="ScreensDesign MCP"` — the answer of a real, running, spec-correct MCP server asking to be authenticated, not a placeholder. The GitHub repo `screensdesign-com/screensdesign-agent-skill` is public, real, and described as "Agent skill for ScreensDesign MCP: mobile app design research for AI agents."

**So the verdict has two halves, not one.** There is no public REST API, no Figma plugin, no export format — `/api` really is a dead end. But there IS a real, working, installable capability: a hosted MCP server, gated behind a paid ScreensDesign login and OAuth, reachable only once a client actually runs the page's JavaScript. **The lesson for next time, recorded here so it does not get re-litigated in three months: a bare `curl` on a modern client-rendered site cannot tell "no route here" apart from "a route that needs the browser to decide" — check an ambiguous route with a JavaScript-executing fetch before calling it a dead end.**

## Patterns worth knowing

- **The story cut, not the slideshow.** One continuous scrubbable video with named, timestamped chapters reads as a flow a person actually lived through, not a folder of stills. Worth the idea even without the tool: when documenting a flow, name its phases and give a time or a step count to each, don't just list screens.
- **Proof travels with the object, not on a separate page.** A revenue figure sits beside every single app card and on every detail page — the claim and its evidence are never more than one glance apart.
- **A four-group way to think about a subscription app's first minutes** — basics, business, onboarding, paywall — is a usable checklist on its own, independent of the tool.
- **Ask, don't filter.** The search bar is phrased as a question ("Show me meditation paywalls with a free trial, sorted by revenue…"), not a set of dropdowns. It treats browsing as something you ask for.
- **The gate shows its own size before it asks to be opened.** A locked thumbnail still shows its timestamp under "Unlock Pro" — you see exactly how much more there is before being asked to pay for it. Withholding content without hiding its existence.

## The lessons for Vox Populi — and the honest limits

**The corpus is a different bet than ours.** ScreensDesign's 2,600 apps are almost all optimising for one thing: turn a stranger into a paying subscriber inside one sitting. Vox Populi is asking a mayor's staff, in front of their own citizens, to trust a process with public opinion. Conversion pressure and trust are not the same currency — some patterns transfer straight, some need real adapting, one whole category should stay refused.

**Copy — proof right where the doubt is.** Every ScreensDesign screen puts a real number beside the claim it makes: "$59M/mo" next to "Claude by Anthropic," a revenue figure on every single app card. The member app's first screen, [voxpopuli.one/app/](https://voxpopuli.one/app/), has no such thing today — a logo square, a title, a paragraph, a large empty gap, then an email field near the bottom. Nothing on that screen tells a citizen this is a real tool their own town already uses before it asks for an email address. The transferable move is not "add a number" for its own sake — it is: put proof of the thing exactly where trust is being asked for. A named, real example (an actual commune, an actual open question) sitting in that gap does for a citizen what the revenue figure does for an app buyer.

**Copy — give the empty gap a job.** ScreensDesign never leaves a space doing nothing — it always shows a number, a name, a chapter or a next step. The console's dashboard, [voxpopuli.one/admin/](https://voxpopuli.one/admin/) ("Démo" tenant), is not broken — it is correct: two projects, two active consultations, six consultations total, two short lists, two buttons. On a big screen it still reads as unfinished, because the space below that real content says nothing. The lesson: on a low-count dashboard, empty space below real content needs a next-step line or a plain state message ("nothing else yet — here is what to add next"), the same way a ScreensDesign screen never just goes quiet.

**Adapt — sort by a trust number, not a money number.** ScreensDesign's whole order is "top-grossing" — money first. A civic tool has its own honest numbers to sort and show by instead: citizens reached, citizens who actually answered, communes already running it. Using THOSE numbers the way ScreensDesign uses revenue is the adapted version of the pattern — money itself, as a headline number on a citizen-facing screen, would read as exactly the wrong kind of proof for this product.

**Refuse — the conversion playbook itself.** Pre-permission explainers written to soften an ask, single-select persona pickers before any value is shown, paywalls dressed as feature lists, urgency countdowns, "join 1M users" testimonial walls — these exist on ScreensDesign because they measurably raise subscription rates. On a tool a citizen is asked to trust with a civic opinion, that same pressure reads as manipulation, not welcome. Refused outright, not softened.

**Refuse — consumer-app density as a bar to clear.** Seeing 2,600 tightly packed onboarding flows makes "fill more of the screen" feel like the obvious fix for anything that looks sparse. Resist it for the console: three honest numbers and two short lists are not thin, they are legible. The actual fix for the empty gap above is one next-step line, not a wall of stat tiles borrowed from a fitness app.

## What we refused to reproduce, and why

No screenshots of ScreensDesign's app grids, teardown videos, or screen-thumbnail strips appear on this page or the linked HTML view. That curation — which apps, which screens, which order — is ScreensDesign's paid product, not the underlying apps' public screens (the rule already stated at the top of [references.md](references.md)). Everything above is a plain-word note on a pattern seen live, cited by URL, dated 08-08-2026. The two images that DO appear on the linked HTML page are our own: live screenshots of `voxpopuli.one/app/` and `voxpopuli.one/admin/` (démo tenant), taken the same day — evidence for the lessons, not decoration.

## Read next

[screensdesign.html](screensdesign.html) — the same note as a reading page, with the two Vox Populi evidence screenshots. [references.md](references.md) — the full inspiration library and its honest-use rules.