Swatchbook MCP: The AI Interface Your Design Tokens Are Missing

The Design Token Gap AI Never Knew It Had

Design tokens are the backbone of modern design systems — but until now, AI agents had no clean way to read them. Swatchbook MCP changes that with a focused, opinionated server that exposes DTCG design token data directly to AI agents, no Storybook instance required.

That's a bigger deal than it sounds.

The Swatchbook MCP server acts as a bridge between your DTCG (Design Token Community Group) token project and any AI agent running on the Model Context Protocol. It surfaces tokens, axes, and diagnostics — the raw material of your design system — in a form that language models can actually reason about.

The key insight here: you don't need Storybook running. This is a headless, agent-friendly interface to your token data. For teams already using the DTCG format (and increasingly, you should be), this is the missing connector piece.

Under the hood, the server parses tokens through Terrazzo — the battle-tested DTCG parser from the Terrazzo team. That means alias resolution, resolver evaluation, and full DTCG compliance come baked in, not bolted on. The choice of Terrazzo as a foundation is a smart one; it's not reinventing the wheel where it doesn't need to.

The current tool surface is lean but meaningful:

• Packages — exposes the package structure of your DTCG project
• Development — built on pnpm workspaces and Turborepo, Node 24, ESM throughout
• Credits — transparent about its Terrazzo dependency, which is a good sign for maintainability
• License — MIT, so you can use it freely without legal headaches

It's worth being honest: this is an early-stage server. The tool count is small, and there are no resources or prompts registered yet. But the architecture — DTCG parsing via Terrazzo, headless operation, MCP-native design — is the right foundation to build on.

With an overall score of 88/100, Swatchbook MCP punches well above its 2-star GitHub count. Let's look at where those points come from — and where the gaps are.

The security score of 29 is the standout. For a tool operating in design system infrastructure — where tokens often encode brand decisions, semantic scales, and system architecture — read-only, diagnostics-focused access is exactly the right posture.

Design system engineers building on DTCG tokens are the obvious first audience. If you're already running Terrazzo for token transformation, adding this MCP server to your toolchain is a low-friction win — your AI assistant can now reason about your actual token structure, not a hallucinated approximation.

AI-assisted design tooling developers should be watching this space closely. The combination of DTCG standardization and MCP connectivity is genuinely new territory. Swatchbook is early, but the pattern it's establishing — headless token exposure for AI agents — is going to matter as design systems become more complex.

Storybook users get an obvious benefit: they can query their design token system without spinning up a full Storybook instance. For CI environments, AI pipelines, or lightweight development setups, that's a meaningful reduction in overhead.

Swatchbook MCP won't win a popularity contest today — 2 stars and a small toolset tell that story plainly. But the 88/100 overall score reflects something real: a tight architectural focus, solid security posture, and a genuinely useful niche at the intersection of design systems and AI tooling.

Design tokens need an AI interface. DTCG is the right standard to build on. And Terrazzo is a solid parsing foundation. Swatchbook MCP has put those three pieces together before most teams even knew they needed it.

Get in early, or catch up later.