Mobbin MCP Server Use Cases: 317k Queries + 30 Prompts

Mobbin MCP Server Use Cases: What 317,427 Real Queries Taught Us

We tracked 317,427 queries in the first month of launching Mobbin MCP. Only 16 said “generate me UI”.

Published

Jul 15, 2026

Cover Illustration

Erik Carter

Author

Rebekah Bek

“Designers are cooked” has been the running headline for over a year now, but by all accounts, they seem more to be on the cooking side of things.

We shipped Mobbin MCP at the end of April. Sounds all tech-y, but really it just lets you plug your favorite AI tool like Claude, ChatGPT, or Cursor into Mobbin’s library of 600,000+ shipped app screens. No more screenshotting apps, pasting them into a doc, and explaining to AI what you’re looking at – just chat like normal and AI pulls design references straight into your conversation. 10,105 people connected it in the first month. 317,427 queries came through.

I’ve been checking them in the mornings. I kept waiting for the big generative prompts — “design me a fintech app and make it not suck”, “generate a dashboard with data viz”. They never showed up.

The number one search, from 319 separate users: ”login screen”. Second: “onboarding welcome screen”. Then:

”dashboard”, “settings screen”, “pricing page”, “empty state”, “checkout flow”.

Not one of the top 50 says “generate”. People are using it the way designers always have: to look things up.

Here's what 10,105 people actually did with it: the six MCP server use cases that kept showing up.

Look up any screen type mid-build

Imagine this: you’re in Cursor, halfway through a settings page. You need to see how other apps handle it — what goes in the sidebar vs. the main panel, where the danger zone sits, whether anyone’s solved nested settings better than tabs. You type something about settings pages and an image gallery of screens shows up in your conversation. You can scroll to see Notion’s settings, Linear’s preferences panel, Stripe’s account page in your chat window.

Set up Mobbin MCP, type in Claude desktop: show me Notion's workspace settings

You can also ask questions about what comes back. Ask which of these use a sidebar vs. tabs, or how they handle the danger zone. The screens are already in the conversation for you to reference, and the follow-up is instant.

This is most of what people do. Onboarding, dashboards, and paywalls account for almost 30% of all queries. After that it’s everything else: navigation, forms, checkout flows, chat interfaces, empty states. If you’ve shipped an app, you’ve designed three of these this month.

Actual MCP queries: “login screen” (319 users) · “dashboard” (48) · “settings screen” (38) · “pricing page” (27) · “empty state” (16)

Prompts to try:

I'm building a login page. Show me how other apps handle it — social login buttons, password recovery link, the copy around the sign-up toggle. What do most of them put above the fold?
I need dashboard layouts for a B2B analytics product. Real dashboards — data tables, sidebar navigation, how they handle chart density.
How do productivity apps handle their settings page? Sidebar vs. tabs, nested settings, where the account deletion option sits. Summarize the most common layout.
Show me empty states that get users to take action — not just 'nothing here yet' with a sad illustration.
How do apps handle bottom sheets? Payment confirmations, date pickers, filter options. What height, what gestures, when do they go full-screen instead?
How do productivity apps handle their color hierarchy? Primary actions vs. secondary vs. destructive. What's the accent, what's the neutral?

Study how a specific app handles it

Many people don’t search by screen type. They search by the app they want to build like — not to copy, but to see how it’s been done.

Linear. 1,268 users searched for it. Notion: 821. Stripe: 674. Duolingo: 546. Airbnb: 474.

About 15% of all queries reference a specific app by name. The other 85% are generic pattern searches like “login screen” or “checkout flow.”

The searches go specific. Not just “Linear” but “Linear dashboard,” “Linear navigation,” “Linear project views.” Same with Duolingo — 546 users across dozens of variations. “Duolingo onboarding.” “Duolingo onboarding flow.” “Duolingo gamification.” For most of these people, Duolingo is a benchmark.

I tried: show me Duolingo’s onboarding flow, first screen to first lesson. Claude pulled the full flow right into the chat as an image gallery: 19 screens before the first lesson. No account creation before lesson one, and sign-up only shows up after you’ve already earned XP. The permissions dialog fires at screen 13, after you’ve already committed to a daily goal.

Actual MCP queries: “duolingo onboarding” (20 users) · “airbnb” (18) · “stripe dashboard” (9) · “headspace onboarding” (9)

Prompts to try:

Show me how Notion handles their workspace setup. Team spaces, sidebar organization, permissions. How do they onboard a new team member vs. a solo user?
Show me how Linear handles their project dashboard. The list view, filtering, and keyboard navigation.
I need to see Stripe's billing and payments screens. How do they present that much data without it looking like a spreadsheet?
How does Airbnb handle search and filtering on mobile? Map view, list view, the filter sheet.
Show me Spotify's mobile navigation — the library, search, and now-playing bar. How do they handle it across different screens?

Search screens for your industry

Search “onboarding” and you get results from every category. Search “banking app onboarding” and you get KYC verification, document uploads, regulatory disclosures.

Fintech is the most-searched vertical at 9.4% of all queries. Most of the top 10 fintech queries are some version of “banking app onboarding.” AI and LLM interfaces are close behind at 7.2%, with designers benchmarking against ChatGPT, Cal AI, Sana AI, solving problems that didn’t exist two years ago: thinking indicators, response streaming, tool-use displays, conversation branching.

Same with food delivery (tipping logic, driver ETAs, real-time order tracking) and fitness (maybe it’s just me, but convincing me to do 45 minutes of cardio is a slightly different problem than an empty inbox).

Here’s me being curious about how fitness apps do just that. I typed “show me a fitness app screen convincing me to do cardio” and the first one back told me optimizing my heart rate zone burns more calories overall. I mean, I’m in.

Set up Mobbin MCP, typed: show me a fitness app screen convincing me to do cardio in Claude desktop.

Actual MCP queries: “onboarding screens from banking apps” (41 users) · “fintech onboarding” (14) · “ai chat interface” (8) · “fintech app onboarding welcome screen” (10)

Prompts to try:

I'm designing onboarding for a neobank. Show me how fintech apps handle KYC verification, document uploads, and trust-building screens. What copy do they use to build trust before asking for bank details?
I'm building an AI chat product. How do other AI tools handle the conversation UI? Message threading, code blocks, thinking indicators, response streaming.
Show me onboarding screens from B2B SaaS apps. How do they explain what the product does in the first three screens?
I'm working on a food delivery checkout. Show me how apps handle tipping, delivery estimates, and order tracking in the same flow.
How do fitness apps handle the first-run experience? Collecting user goals, fitness level, recommending a plan.

Prove a design decision to your team

206 designers searched three or more different apps in a single session.

You’re in a review. Your PM pushes back on a layout. Instead of defending it from gut, you pull screens from five apps that handle it the same way: here’s Linear, here’s Stripe, here’s Notion. That’s the convention. A junior designer with reference from six shipped products is harder to argue with than a senior designer with a feeling.

It’s also changed how teams align. Instead of spending an hour prepping examples for a meeting, you surface real screens mid-conversation so discussions get concrete immediately.

This isn’t only a designer thing, by the way. If you’re a writer or marketer like me, pull up how other apps handle their copy and messaging. What they lead with, how they frame free vs. paid, whether they list features or just show the price.

To go deeper, compare competitors side by side. While one app shows you one solution, three apps solving the same problem the same way shows you a convention you probably don’t want to break. And if no one’s doing it your way, better to find out now than in the review.

The combinations people actually searched are all direct rivals. Calm and Headspace. Oura and WHOOP. Linear and Notion. ChatGPT and Claude. Monzo and Revolut. Nobody pulls up Calm and Headspace in the same sitting for vibes. They’re comparing how each app handles session selection, subscription pricing, or the meditation player screen.

Personal finance has a three-way cluster: Copilot Money, Monarch, and YNAB. All three in the same session — budgeting, transaction categorization, spending summaries. Sure sounds like someone prepping for a design review.

Top co-searched pairs: Calm + Headspace (15 users) · Oura + WHOOP (13) · Linear + Notion (12) · ChatGPT + Claude (9) · Monzo + Revolut (9)

Prompts to try:

Compare how Calm and Headspace handle session selection. Categories, duration, the play screen. Side by side.
Pull up daily summary screens from Oura and WHOOP. How does each one present sleep and recovery data?
How do Linear and Notion handle project navigation differently? Sidebar structure, view switching, moving between projects.
Compare transaction screens in Monzo, Revolut, and Wise. How does each handle receipts, categories, and spending breakdowns?
Pull up ChatGPT, Claude, and Perplexity. How does each one handle conversation history, the input area, and settings?

Build it into your workflow

Someone in Ukraine built an agent that pulls Mobbin reference for every Jira ticket before anyone starts designing. The agent reads the ticket, figures out what kind of screens the task needs, and pulls 20 reference screenshots automatically. The queries arrive in English and Ukrainian, tagged with the product name and task — across six products simultaneously. Nobody needs to type anything; the references are just there when they open the ticket.

They’re not alone. 1,255 users (12.4% of everyone) ran 50 or more queries in the month. 29 users ran over 500. The heaviest real user (not an automated scraper) averaged 247 queries a day for 18 straight days. Another ran 121 a day for 26 consecutive days.

They folded reference into everything else: writing code, filing tickets, running design reviews. The most natural pairing is Mobbin to Figma — pull reference, analyze it in Claude, then build the moodboard in Figma without leaving the conversation. Same thing works with Paper.

A look at the outcome:

Actual MCP query from a power user: “Find 20 real Mobbin mobile screenshots for Jira task: ‘A/B Meeting bot’. Product: Notee.”

Prompts to try:

I'm about to start designing three screens: login, onboarding, and a dashboard. Before I open Figma, pull reference for all three from apps with strong UX.
[paste your brief or ticket] Based on this brief, pull reference screens from apps solving similar problems.
I have three meetings today where I'm presenting design options. Pull reference for a notification center, a scheduling picker, and a profile page. All three, before I start.
I'm starting a new project. Here's the PRD: [paste doc]. Pull reference screens for every user-facing flow mentioned in it.
Find dashboard layouts from analytics products. Then create a moodboard in Figma with the best examples.

Ground AI output in what’s shipped

What if you do want to generate? If you’re using AI to build UI in Cursor, Claude, or anything that writes code, you can feed it real screens before you ask it to produce anything.

Without reference, you get placeholder copy (“Welcome back!”) and layouts that look correct but follow no particular convention. One Mobbin user called it “avoiding the very obvious AI-designed UI look.” That’s what grounding does; it keeps the output from looking like something an AI agent guessed at. Feed it real screens first and the output gets specific: loading states, retry paths, disabled controls, minimum order rules, stale data handling, danger actions. All the stuff you only notice when it’s missing.

I asked Claude to build me a settings page for a B2B productivity app, first with no reference, then after pulling settings screens from Notion, Linear, and Stripe and following those patterns.

The first version looked like every AI-generated settings page: white cards stacked on grey, generic blue accent, font-weight 700 on everything, no information hierarchy beyond card = section.

The second time, Claude looked at the references first. It borrowed Linear’s persistent sidebar with a 2px left-border active state, Stripe’s red-separated danger zone, and wrote setting descriptions that say what a setting does (“Shows up in emails and shared links”) instead of what it is (“Visible to all members”).

Zero of the top 50 MCP queries ask the AI to create anything. But when people do generate, the reference is what keeps the output from being one step shy of a hallucination.

Prompts to try:

I need a notification center. Look up how 5 apps handle notification lists — read/unread states, grouping, action buttons. Then generate a notification component using those patterns.
Search for checkout flows from food delivery apps. Then build a checkout component that follows those patterns — tipping, ETA display, order summary.
Search for profile editing screens from social and productivity apps. Avatar upload, bio fields, privacy toggles. Then build a profile editor that follows those conventions.
Look up product search pages from shopping apps — autocomplete, filter chips, recent searches. Then build a search component that handles the empty state, loading, and results.
Look up how meditation apps handle their session player — timer, ambient sounds, progress. Then build a player component that follows those patterns.

The weird queries

None of this is technically useful, but 317,427 queries is a lot, and this was pretty interesting.

53 users searched in Chinese and English at the same time, mapping entire e-commerce flows in a single query: product home, product list, cart, order confirmation, payment success, mini-program. Half in Chinese characters, half in English keywords. Korean showed up too. Japanese. The best one: “ZOZO ZOZOTOWN おにり wishlist fashion Japan,” sitting in the logs next to queries in Mandarin and Korean.

Then there were the AI agents that tried to describe an entire app in one shot, cramming every possible aspect of a screen into a single query until they hit the 500-character limit: “podcast library feed list of episodes audio tracks with player controls progress bar duration metadata actions download delete queue favorites history explore c—” Cut off mid-word.

And 18 people searched “test.”

I still check the queries on and off in the mornings. The list still hasn’t gotten more exciting than “login screen.” I’ll write a part 2 if it does.

Mobbin MCP by the numbers

Login screen

319

users

#1 search: "login screen" (319 users)

MCP queries from 10,105 people in one month.

317,427

#1 app (1,268 users)

Linear

daily active users by June 2026

~1,200

out of 317,427 queries asked the AI to generate a design

#1 industry (9.4% of queries)

Fintech & banking

How to use Mobbin MCP

Setup takes under a minute on any paid plan. See the step-by-step guide on the Mobbin MCP page, and per-client install guides for Claude Code, Claude Desktop, Cursor, and Codex are in the docs. Once you're connected, every prompt in this study works as written.

The UI MCP stack: what else to plug your AI into

Mobbin MCP is one piece of a bigger picture. If you're wiring up AI tools for design work, these are the UI MCP servers designers actually combine, and the job each one does. (Same move as the intro, but for the rest of the stack.)

Mobbin MCP: design inspiration MCP and reference. 621,500+ shipped screens your agent searches for real patterns. (Yes, ours; the 317,427 queries above are the evidence.)

Figma MCP: your own design files to code. The execution half of the workflow.

Design system MCPs: a design system MCP like Supernova or southleft gives agents your components and guidelines, and a design tokens MCP server keeps colors and spacing consistent across everything the agent builds.

UI component MCPs: generated UI design MCP building blocks from magicui or shadcn when you need net-new components fast.

Which design tools support MCP changes monthly, but the pattern holds: an MCP for designers is either reference (what works), system (your rules), or generation (an MCP for UI design making new parts). The best design MCP setup uses one of each; every AI design MCP workflow in our data leans on reference most, which is why the six use cases above look the way they do.

The short version

What are MCP server use cases?

Most MCP use cases you read about are hypothetical. These six came from 317,427 real queries: looking up screen types mid-build, studying specific apps, searching by industry, proving design decisions, workflow integration, and grounding AI output in shipped designs. Not one "generate" in the top 50.

Is Mobbin MCP free?

Mobbin MCP is included in every paid Mobbin plan; there is no separate MCP fee. It is not on the free plan. Details on the Mobbin MCP page.

How do you connect it to Claude or Cursor?

One command, no API key. The per-client steps (Claude Code, Claude Desktop, Cursor, Codex) are in the docs; setup takes under a minute.

Which design tools support MCP?

Figma, Mobbin, Supernova, magicui, shadcn and a growing list all ship MCP servers. See the UI MCP stack section above for what each one is for.