TrademarkTrademarkDocs
Ctrl k
Search docs...
Docs

AI Coding (MCP)

Updated: July 5, 2026
View on GitHub
Thumbnail of AI Coding (MCP)
Once UI is an agent-first design system that empowers indie creators to build high-quality, responsive apps — without wrestling with complexity. Designed for AI-native workflows, it turns abstract intent into concrete interfaces with minimal syntax. For writing UI code, use the harness — not full doc pages. It keeps context under ~10KB per task.
Artifact
URL
When to load
Manifest/ai/manifest.jsonEntry point — lists all harness files
Compact rules/ai/rules.compact.mdAlways, before any UI task (~2KB)
Component catalog/ai/catalog.jsonPick components by purpose/tag
Task bundle/ai/tasks/index.jsonMatch user intent → tasks/pricing.json, etc.
Component slice/ai/components/Grid.jsonOn demand — one file per component
Gold example/ai/examples/*.tsxMimic structure for the task type
Pro blocks/ai/examples/blocks/manifest.jsonProduction-quality composition references
Gotchas/ai/gotchas.jsonWhen using Background, Card, RevealFx, Fade
npm imports (version-pinned): Machine-readable discovery: onceUi field in @once-ui-system/core package.json. Workflow:
  • Load rules.compact.md
  • Resolve intent → pick a file from tasks/index.json (e.g. tasks/pricing.json)
  • Fetch component slices listed in the task bundle
  • Read the gold example for structure; for production fidelity, read matching Pro blocks from examples/blocks/manifest.json
  • Generate code
  • Validate: pnpm validate-ai-code path/to/file.tsx
For how does X work? questions, agents can pull docs through Context7's MCP server. To configure it:
  • Open mcp_config.json in File / Preferences / Settings / Plugins
  • Add the following configuration:
  • Restart your IDE
  • Run: get-library-docs /once-ui-system/core
Use Context7 for exploration. Use the harness for codegen. Compared to Tailwind + shadcn, Once UI yields up to 70% more compact output, with cleaner, more natural syntax — semantic Row/Column, token props, and purpose-built components. That compactness only pays off when agents load the harness and follow the rules — especially layout shorthands (fill, center), surface recipes, and component appropriateness (Card is interactive-only).
Edit this page on GitHub
TrademarkTrademark
Built with curiosity by Lorant One and the Design Engineers Club