Answers with evidence
Ask “how does a paywall frame its trial?” and get the screens, the flow they belong to, what components they use and the project each one came from.
find_ui_references
MCP server · runs locally over stdio
Interface references your agent can actually use.
ThinkUI plugs into the coding agent you already use and answers design questions with evidence: the screen, the flow it belongs to, the component itself, the typeface. Every result says where it came from, so you can keep building instead of second-guessing.
470written design references
3,802ready-made components
1,698animated icons
118fonts, icon sets and systems
Measured from the library this server serves — not projections. Ask the server for corpus_stats and it reports the same numbers.
What it does
Design decisions, with the reasoning attached.
An agent that improvises a paywall produces a paywall nobody has ever shipped. ThinkUI gives it something real to work from — and tells you what that something was.
Components you can paste
Buttons, loaders, cards, forms, inputs, switches, tooltips and patterns — returned as the component's own markup and stylesheet, not a description of one.
find_ui_elements
Motion that actually moves
Animated icons arrive as working SVG — animation included, no JavaScript runtime, no icon font, nothing to install. Paste it and it animates.
find_animated_icons
Type and systems
The right typeface for a brief, an icon set that is still maintained, the design system a product like yours is built on — with the state of each project attached.
find_ui_materials
It refuses when it should
Asked for a navigation bar or a page footer, it returns nothing and tells you what it does hold. A confident wrong answer costs more than no answer.
no invented results
Fits your setup
A local process your agent starts. Add one block to its config and the tools appear in every conversation you have in that client.
MCP over stdio
How it works
Three steps, once.
Connect it
Paste one block into your agent's MCP config. No keys to rotate daily, no build step, nothing installed into your project.
Ask in plain language
“Design the onboarding screen for a budgeting app.” The agent reaches for references instead of inventing them.
Build from something real
You get screens to follow, components to paste and assets to ship — with a trail back to where each came from.
One block of config.
ThinkUI ships as a standard MCP server. Add it to Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed or any framework with an MCP client — the same block works everywhere, and it is the only setup step.
Your account key goes in one place. Everything else — tools, prompts, transport — is handled by the server.
Full quickstart{
"mcpServers": {
"thinkui": {
"command": "npx",
"args": ["-y", "thinkui-mcp"],
"env": { "THINKUI_KEY": "your-key" }
}
}
}
Then ask your agent something you would normally answer by opening ten apps.
Pricing
Same server, same tools. Pay for the volume you use.
Start free. Move to a paid plan when it is carrying real work — you can change plan or cancel from your account at any time.
$0personal use
For trying it on a side project.
- 200 tool calls a month
- All five tools
- Components and icons
$19per month
For work you are actually shipping.
- Unlimited tool calls
- Commercial use, including client work
- Screens returned as images
- Priority when new sources land
$49per month
For a small team on the same design language.
- Five seats
- Shared saved queries
- Everything in Pro
Prices in USD. Compare plans in detail →
Questions
The ones people ask before connecting it.
What is MCP?
The Model Context Protocol is the standard way an AI coding agent connects to an outside tool. You add a server to your agent's config once, and its tools become available in every conversation you have in that client.
Which clients does it work with?
Any client that speaks MCP over stdio — Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed, and agent frameworks with an MCP client. The config block is the same everywhere.
Does anything about my project get sent anywhere?
No. The server runs as a local process that your agent starts, and it answers from its own reference library. Nothing about your code, your files or your project is involved in producing a result.
What happens when it does not know something?
It returns nothing and tells you what it does hold. Ask for a navigation bar or a page footer and you get an empty result with the categories that exist, rather than the nearest component that happens to share a word with your question.
Can I use what it returns in commercial work?
Yes on a paid plan. Every result the server returns identifies where it came from, so your own records stay in order when you ship.
How do I cancel?
From your account page, in one click. Cancelling stops the next charge; your key keeps working until the end of the period you already paid for.
Stop describing interfaces to your agent. Show it some.
Connect ThinkUI and give it something real to work from.
Free tier available · Cancel any time · Runs locally in your agent