The MCP server below is a design spec, not a shipped product. Implementation lives in a separate repo (gunjo-ui-mcp) and will land after the 1.0 stable spec freeze. Until then, use the Copy-spec-for-AI manual workflow.
Track progress in the issue tracker.
gunjo-ui MCP server
A Model Context Protocol server that lets Claude, Cursor, and other MCP-aware clients query the GunjoUI namespace directly — no copy-paste of specs needed.
Why MCP
MCP turns the Phase 5 spec endpoints into a typed, discoverable namespace. Instead of pasting Markdown, users say "use GunjoUI" and the AI tool callslist_componentsandget_component_spec as needed. Closes the loop: AI can ask for what it needs at the moment it needs it.
Planned tool surface
list_components
{ category?: "inputs" | "display" | "feedback" | "navigation" | "overlay" | "layout" | "patterns" }Component[] — title, description, category, slug for each
“Show me all template components”
get_component_spec
{ category: string, name: string }ComponentSpec — full Markdown + JSON spec for one component
“Pull the spec for display/data-table”
get_design_tokens
{ group?: "color" | "typography" | "spacing" | "shadow" | "radius" | "animation" }Token[] — name, value, css-var, theme variants
“List all color tokens”
search_components
{ query: string, limit?: number }Component[] — fuzzy-matched against title and description
“Find components for showing loading state”
Architecture
The MCP server is a thin CLI that proxies to the existing/api/specsendpoints, normalizing responses into MCP tool results. No state of its own — the docs site remains the single source of truth.
# Install (when shipped) npm install -g gunjo-ui-mcp # Run as a Claude Desktop / Cursor MCP server npx gunjo-ui-mcp \ --base-url https://gunjo.dev \ --transport stdio
Until the MCP server lands
The Markdown "Copy spec for AI" button on every component page already covers the manual workflow. Drop into v0 or Cursor — same fidelity, just a copy-paste away.
Use the manual workflow