Design Systems Mcp

What is Design Systems MCP Server?

An AI-powered Model Context Protocol (MCP) server providing intelligent access to authoritative design systems knowledge. It is powered by Supabase vector search with 200+ curated entries including W3C standards, WCAG guidelines, and design system best practices. It is intended for developers, designers, and accessibility specialists who need quick, accurate answers about design systems.

How to use Design Systems MCP Server?

Connect any MCP-capable client (e.g., Claude Desktop, Cursor, Windsurf) to the public endpoint https://design-systems-mcp.southleft.com/mcp with zero setup. For local development, clone the repository, run npm install, configure environment variables, and start with npm run dev. The server exposes four MCP tools: search_design_knowledge, search_chunks, browse_by_category, and get_all_tags.

Key features of Design Systems MCP Server

• Production vector search using Supabase pgvector and OpenAI embeddings.
• 200+ curated entries from W3C, WCAG, ARIA, and 10+ major design systems.
• Hybrid search combining vector similarity with keyword matching (0.15 threshold).
• Edge-optimized deployment on Cloudflare Workers with global distribution.
• Streaming responses via SSE for fast time‑to‑first token.
• Source reliability badges marking each answer as Primary/Authoritative/Reference/Example/Community.

Use cases of Design Systems MCP Server

• Quickly look up WCAG 2.2 color contrast requirements while designing.
• Retrieve W3C Design Tokens specification for token naming.
• Browse component patterns from Material Design 3 or IBM Carbon.
• Verify ARIA Authoring Practices for accessible UI patterns.
• Get best‑practice guidance from multiple design systems in one query.

FAQ from Design Systems MCP Server

What clients does Design Systems MCP Server support?

It works with any MCP-capable client, including Claude Desktop, Claude Code (CLI), Cursor IDE, Cline (VSCode), Continue (VSCode), Zed Editor, and generic MCP clients.

How do I add my own content to the server?

You can ingest web content via npm run ingest:url <URL>, bulk ingest from CSV with npm run ingest:csv <file>, ingest PDFs with npm run ingest:pdf <file>, or crawl entire websites with npm run crawl:website <URL>. After ingestion, run npm run ingest:vectors to generate embeddings.

Is there a live demo or hosted web UI?

Yes. The live demo is at https://design-systems-mcp.southleft.com/. It includes an AI chat interface that streams responses via SSE and a landing page with MCP endpoint details.

What MCP tools are available?

Four tools: search_design_knowledge (semantic search across the knowledge base), search_chunks (find specific content in chunks), browse_by_category (browse content by categories like components, tokens, patterns), and get_all_tags (list all content tags for filtering).

Does the server require authentication or API keys?

The public MCP endpoint is available without authentication. Local development requires setting environment variables (e.g., Supabase and OpenAI keys) in a .dev.vars file.