Skip to main content
PET BRAINS

// tool · 03 / 03

Designlib-MCP

Curated design-knowledge catalog for AI coding agents. 400+ inspiration pages, palettes, fonts, and patterns — over MCP.

View on GitHub ↗

// what is it

A Model Context Protocol server for design knowledge.

Designlib-MCP is a Model Context Protocol server that gives any MCP-capable AI coding agent — Claude Code, Cursor, Copilot, Cline, others — direct access to a curated design-knowledge catalog. 400+ inspiration pages, hundreds of palettes, font pairs, icons, animations, landing patterns, chart types — all addressable as tools the agent can call.

Unlike a Figma plugin or a static reference page, designlib-mcp speaks the agent's protocol natively. The agent fetches what it needs at the moment it needs it. Open source, MIT-licensed, hosted free.

// the problem

AI agents design from training data, not from references.

Most AI coding agents have seen a million Dribbble shots and zero of your brand references. The result is well-trodden defaults: cyan-on-dark, gradient text, three-column card grids — the visual mode the training data over-represents.

Designlib-MCP fixes the input side of that equation. Instead of relying on the agent's averaged-out memory of design, the agent fetches actual references — palettes resolved by mood and contrast character, inspiration pages anchored by structural axis, font pairs verified for license — at the moment it generates output.

The agent stops guessing what "warm editorial" looks like. It looks it up.

// how it works

Connect the server. Call the tools.

  1. // 01

    Add the MCP server to your agent.

    One config line in your agent's MCP settings — Claude Code, Cursor, Copilot all support the same protocol. The server runs hosted; no install, no API keys for the catalog.

  2. // 02

    The agent calls catalog tools.

    list_palettes, get_inspiration_page, list_animations, get_font_pair, plus facets for filtering. Singular filters per call, multi-call dedup pattern. The agent learns the API the way it learns any tool.

  3. // 03

    Output anchored to real references.

    Whatever the agent produces — Figma frames, code, design specs — is anchored to actual source material from the catalog, not training-data averages. Source IDs surface back so you can audit what was used.

// who it's for

Builders running design pipelines through AI agents.

Designlib-MCP is for anyone whose AI agent generates design output — UI code, Figma frames, design tokens, brand systems. Indie founders shipping single-file landings, agencies running production design with AI, design-builder users (the plugin pulls from this catalog).

Best fit: agents that already speak MCP (Claude Code, Cursor, Copilot, Cline). Worst fit: closed-tool workflows where MCP isn't supported — there's no client there to connect.

// compatibility

Compatibility

Protocol
Model Context Protocol (MCP)
Clients
Claude Code, Cursor, Copilot, Cline, any MCP-capable client
Transport
HTTP, hosted at designlib-production.up.railway.app/mcp
Catalog size
400+ inspiration pages, 100+ palettes, font pairs, animations, icons, patterns
License
MIT
Status
v1 shipped, public

// get started

Add one MCP config line.

Open source, MIT-licensed. Hosted free at designlib-production.up.railway.app/mcp — or self-host the server from the repo.

View on GitHub ↗

Get notified when Designlib-MCP ships a major update.