Use the design library from ChatGPT, Claude, Cursor, and other AI tools

What is MCP, in one paragraph

MCP (Model Context Protocol) is an open standard that lets AI assistants use external tools — the same way a browser extension extends your browser. This library runs a hosted MCP server, so any MCP-capable assistant (ChatGPT, Claude, Cursor, and others) can search the catalog, read the written descriptions and tags, pull up similar references, and work with color palettes. You connect it once; after that it's just part of your assistant's vocabulary.

What this lets you do

What your assistant can retrieve

• Search the full catalog with natural-language queries plus structured filters (style, color, typography, mood, industry)
• The complete written description of any reference: layout, hierarchy, type treatment, color use, and the key takeaway
• Similar references to any item
• The library's facet vocabulary, for building precise queries
• Color palette tools: analyze a palette, generate one, suggest pairings

Which tools work

ChatGPT (as a connector), Claude (claude.ai, Claude Desktop, Claude Code), Cursor, and any other MCP client. The server lives at https://mcp.vasa.works/mcp.

How to connect

One server URL, a few ways in. ChatGPT and Claude sign you in through the browser; key-based clients paste the personal API key from your account page.

ChatGPT

1. Open Settings → Apps & Connectors and create a new connector.
2. Paste https://mcp.vasa.works/mcp as the server URL and choose OAuth.
3. Sign in with your email when ChatGPT asks, then enable the connector in a chat.

Claude

1. Open Settings → Connectors and add a custom connector.
2. Paste https://mcp.vasa.works/mcp and approve the sign-in in your browser. No key pasting.

Claude Code

Add the server from your terminal:

claude mcp add --transport http inspiration https://mcp.vasa.works/mcp

Then run /mcp in Claude Code, pick inspiration and sign in in your browser.

Cursor and other key-based clients

Generate your API key on the account page, then add the server to your MCP config:

{
  "mcpServers": {
    "inspiration": {
      "url": "https://mcp.vasa.works/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Then ask your assistant for design references. That's it.

Agent skills

Skills teach your coding agent how to use the library well: when to search, how many calls to spend, and how to turn references into work of your own. They are open source and use the same MCP server.

Claude Code plugin

Run these in Claude Code. The plugin connects the MCP server for you, so you can skip the step above.

/plugin marketplace add huseyingayiran/inspiration-skills
/plugin install inspiration@vasa-works

Cursor, Codex and other agents

Install the skills with one command, then connect the MCP server as described above:

npx skills add huseyingayiran/inspiration-skills

What the skills do

• inspiration-art-direction: Researches references for your brief, derives a written art direction from them, then builds against that direction and critiques the result.
• inspiration-brand-kit: Builds a starter brand kit from a hex color, a brand name or a few adjectives: a color palette, a type pairing and contrast-checked design tokens.
• inspiration-references: Finds references, compares them side by side and saves the keepers to your boards.

Free vs Pro

Connecting is free. A free account includes 50 tool calls per month — browsing the library's vocabulary doesn't count against it — and Pro is unlimited. When a free quota runs out, your assistant gets a friendly note with the reset date, not an error. Plans on the pricing page.

Privacy

The server receives only the tool calls your assistant makes — search queries, item ids, palette values — tied to your API key for metering. Your conversation itself never reaches us; it stays between you and your AI provider.

Common questions

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI assistants use external tools — the way a browser extension extends a browser. Inspiration by VASA Works runs a hosted MCP server, so MCP-capable assistants like ChatGPT, Claude and Cursor can search the design library, read its written descriptions and tags, fetch similar references, and work with color palettes from inside a conversation.

Is the MCP free to use?

Connecting is free. A free account includes 50 tool calls per month, and browsing the library's facet vocabulary doesn't count against the quota. Pro accounts have unlimited MCP access. When a free quota runs out, the assistant receives a friendly note with the reset date rather than an error.

Which AI tools can connect to the library?

Any MCP client: ChatGPT (as a connector), Claude (claude.ai, Claude Desktop, Claude Code), Cursor, and others. The server lives at https://mcp.vasa.works/mcp. ChatGPT and Claude connect with a sign-in in the browser (OAuth); clients like Cursor use a personal API key from your account page.

What data does the MCP server see?

Only the tool calls the assistant makes — search queries, item ids, palette values — tied to your API key for metering. The conversation itself never reaches the server; it stays between you and your AI provider.

Can the AI design something from the references?

Your assistant can use the references and their written descriptions as context for its own work — layout analysis, direction writing, its own generations. The library supplies the references and the vocabulary; what the assistant builds from them is up to the assistant.

Are there ready-made skills for coding agents?

Yes. The open-source inspiration-skills package adds three agent skills: art direction, brand kit and references. In Claude Code, install it as a plugin with /plugin marketplace add huseyingayiran/inspiration-skills and /plugin install inspiration@vasa-works, which also connects the MCP server. For Cursor, Codex and other agents, run npx skills add huseyingayiran/inspiration-skills.