Netspaces · 2024–25

The AI generating your interface has never seen your design system.

I built one it can read.

Role Head of Design
Focus Design Systems · AI Workflow · Design Engineering

The gap

Anyone can generate a screen now. Claude, Cursor, Lovable. The output looks plausible, and it drifts, because the model has no idea a design system exists.

AI raised the floor of interface creation. It did not replace the ceiling. Consistency was the cost nobody priced in: every screen starts from zero, and the drift compounds. The gap is structural. No prompt closes it.

The move

Put the design system where the model can read it.

An MCP server now sits over the live design system. Any MCP client that generates interfaces queries the components as they ship and the tokens with their real values, instead of guessing.

MCP · design-system server ds.netspaces.org
 get_design_tokens { "brand": "bloco" }

{
  "--foreground":      "#1e2333" 
  "--primary":         "#6078b9" 
  "--primary-hover":   "#4f66a6" 
  "--primary-pressed": "#3f5593" 
  "--highlight":       "#6078b9" 
  "--ring":            "#6078b9" 
  "--display-font":    "Instrument Sans"
  "--book-font":       "Instrument Sans"
}
get_design_tokens, one of nine tools the server exposes. The values are the Bloco brand's, as they ship.

The system underneath

Built to be read. By people and by machines.

It works because of how the system ships. The registry at ds.netspaces.org serves source: run npx shadcn add and the component's actual implementation is copied into your project. What makes that pleasant for a developer is exactly what makes it legible to a model. Nothing hides in a compiled bundle.

Underneath sit two token layers: primitives shared by every brand, and semantics resolved per brand by one data-brand attribute. 56 components and 12 ADRs record how and why. The Figma library stays in sync with the code through the Figma MCP.

Same component.

Every color on this card comes from the brand's tokens.

<section data-brand="netspaces">

--primary
#558eff
--highlight
#ff3e7d
--foreground
#071d49
--display-font
Manrope

Rendered live on this page from the system's token values. Netspaces and Propriedade Digital share a primary; the highlight sets them apart.

Proof in production

Seven landing pages went to production through this pipeline.

This one is live: Formação em Tokenização Imobiliária, on the Propriedade Digital brand, one of the four brands the token system resolves. The proof runs on the system you just watched switch.

Full scroll of the Formação em Tokenização landing page: hero with headline and calls to action, market context with figures, program positioning and course modules, all in the Propriedade Digital brand.
One of seven · lp.propriedade.digital · the live page. It scrolls with you.
The same landing page at phone width: Propriedade Digital header, stacked headline and a full-width blue call to action.
Phone width, same tokens.

Proof in depth

The same pipeline, across an entire product.

The same pipeline generated the 26 screens of an entire product, Go Index, inside its own brand theme. Go Index is internal and has not launched. The landing pages prove breadth across published pages; this is the system holding when the surface is a whole product.

  • Go Index · internal screen
  • Go Index · internal screen
  • Go Index · internal screen

Not shown: Go Index is internal and has not launched.

Where it stands

Where it stands

The skill and the MCP server already work, in production use. A fully consolidated pipeline does not exist yet. That is the current work: consolidating what already works into one.

The point of a higher floor is what it buys back: less time catching drift, more time on the judgment calls only a designer with repertoire can make. The system does not replace that judgment. It holds one designer's standards across every surface an AI touches.