CATCHUI
MCP SERVER

Ask your AI assistant what any UI element is called.

CatchUI's curated pattern catalog speaks the Model Context Protocol. Add one URL to Claude, ChatGPT, Cursor, or any MCP client, and your assistant answers "what is this element called" from 28 curated UI patterns — in English and Chinese, citing canonical teaching cards.

https://catchui.com/api/mcp

Connect your client Browse the catalog first

At a glance

Access

Free. No account, no API key, no authentication — paste the URL and it works.

Transport

Streamable HTTP (stateless JSON-RPC). Works with MCP clients that support Streamable HTTP.

Data

Read-only lookups against the public curated catalog. No user data is touched.

Languages

Queries and answers work in English and Simplified Chinese, including real search phrasings.

Connect your client

Claude Code

Run once in your terminal:

claude mcp add --transport http catchui https://catchui.com/api/mcp

claude.ai / Claude Desktop

Settings → Connectors → Add custom connector → paste https://catchui.com/api/mcp. No authorization step appears because none is needed.

ChatGPT

Settings → Apps & Connectors → enable Developer mode → Create: server URL https://catchui.com/api/mcp, authentication "None".

Cursor and other clients

Any client accepting a Streamable HTTP URL works. Cursor's ~/.cursor/mcp.json:

{ "mcpServers": { "catchui": {
    "url": "https://catchui.com/api/mcp" } } }

Three tools, one job

search_ui_patterns

Turn a plain-language description or question — "the bar at the bottom of an app", 「三条横线的图标叫什么」 — into pattern names, definitions, matching Q&A, and catalog links.

get_ui_pattern

The full bilingual teaching entry for one pattern id: definition, pronunciation, aliases in both languages, common questions, and related patterns.

list_ui_patterns

Browse all 28 patterns grouped by family — navigation, input, action, overlay, feedback, content, and data.

Try asking

What it deliberately does not do

Screenshot recognition is not exposed over MCP — it runs in the web app at catchui.com. Multimodal assistants can still look at your screenshot themselves, describe an element in words, and search the catalog for its proper name.

Frequently asked questions

What is the CatchUI MCP server?

A public Model Context Protocol endpoint at https://catchui.com/api/mcp. Once added to an MCP client such as Claude, ChatGPT, or Cursor, the assistant can search CatchUI's curated catalog of 28 UI patterns and answer what an interface element is called, with definitions and canonical links.

Is the CatchUI MCP server free?

Yes. The catalog tools are free, require no account, no API key, and no authentication. Screenshot scanning is a separate feature of the web app and is not part of the MCP server.

Which apps can connect to it?

Any Model Context Protocol client that supports the Streamable HTTP transport: Claude Code, claude.ai and Claude Desktop custom connectors, ChatGPT developer-mode connectors, Cursor, and others.

What tools does it expose?

Three read-only tools: search_ui_patterns finds pattern names from plain-language descriptions or questions in English or Chinese, get_ui_pattern returns the full bilingual teaching entry for one pattern, and list_ui_patterns browses the catalog by family.

Can it analyze screenshots over MCP?

No. Screenshot recognition runs in the CatchUI web app at catchui.com. Multimodal assistants can still look at a screenshot themselves, describe an element in words, and use search_ui_patterns to find its proper name.

Does it access my data?

No. Every tool is a read-only lookup against the public curated catalog. The server requires no sign-in and touches no user accounts, libraries, or uploads.

Operator and contact

CatchUI is operated from Hong Kong. Questions about the MCP server can be sent to support@catchui.com.

Facts on this page last verified: 2026-08-24.