What is a Design System MCP?

A Model Context Protocol (MCP) gives AI agents on-demand access to external tools and structured knowledge. A design system MCP codifies your components, tokens and guidelines into machine-readable metadata so LLMs can query and generate UI code that conforms to your system. This guide is based on real talks from 4 speakers at the AI Design Systems Conference 2026 by Into Design Systems: Diana Wolosin (Indeed, with her 1056-prompt benchmark), Jesse Gardner (New York State), Laura Fehre (Figma) and Romina Kavcic (The Design System Guide). All 15+ hours of conference recordings are available at intodesignsystems.com.

Key concepts

  • AI is a new user (Diana Wolosin, Indeed): Your design system was written for humans. The new user is AI, and it needs structured metadata, not documentation prose.
  • MCP deterministic, LLM stochastic (Diana Wolosin, Indeed): Same MCP query returns the same data every time. Same MCP results sent to the LLM produce different reasoning each time. Structured data mitigates that randomness.
  • JSON for structured data, Markdown for rules (Diana Wolosin, Indeed): JSON for MCP component APIs (props, sizes, variants): it's a contract. Markdown for natural-language instructions and rules for the LLM.
  • Progressive disclosure of context (Diana Wolosin, Indeed): Always-on rules for foundations (spacing, color, typography), MCP on-demand for components, AGENTS.md as orchestration layer. Together they form a plugin.
  • Foundations must be always-on (Diana Wolosin, Indeed): MCP only returns what you asked for. If the prompt says 'build me a card,' the LLM fills gaps with assumptions. Spacing and color belong in always-on rules.
  • Guidelines are not laws (Laura Fehre, Figma): Writing more rules into a guideline file doesn't control the outcome. In nearly 100% of cases the prompt will win over the guidelines.
  • Don't just plug MDX into an MCP (Diana Wolosin, Indeed): Connecting human documentation to an MCP is 5x more expensive and less accurate than structured JSON. Diana's benchmark proved this on Indeed's real design system.
  • Code as source of truth for MCP (Jesse Gardner, New York State): JSDoc on Lit web components feeds the MCP with props, types, and usage guidance. Design and code stay in sync through a single structured source.

Diana's MCP benchmark findings

  • Setup: Cursor plus Claude Sonnet 4.5 plus Indeed's design system MCP, stored in Vectra (open source vector DB). 22 prompts run 3 times each across 8 MCP configurations for a total of 1,056 prompts.
  • Formats tested: Markdown MDX (the original human docs), plain Markdown, hybrid Markdown + JSON, JSON (baseline and variations), and TOON (Token-Oriented Object Notation designed for AI).
  • Metrics: Input quality (cost, average tokens, relevance, query coverage) and output quality (coverage, hallucinations, average props found).
  • Finding #1: MCP is deterministic, LLM is stochastic. Same input to MCP returns the same output. Same MCP results sent to the LLM produce different reasoning each time.
  • Finding #2: Structured data mitigates AI uncertainty. Hybrid (MD + JSON) showed the most variation across runs and was unreliable. JSON showed the most consistency.
  • Finding #3: JSON has the highest accuracy when verified against actual documentation, and uses 80% fewer tokens than hybrid Markdown for equal or better accuracy.
  • Winner: JSON. Sharpest responses at the lowest token cost. Diana built this benchmark in ~3 months, vibe-coded with Cursor.

Real-world implementations

  • Indeed (Diana Wolosin): 77 components processed via JavaScript parsers (one per knowledge domain: accessibility, development, localization, design) generating structured JSON. Original MDX in GitLab → parsers built with Cursor → JSON → ingestion into Vectra vector DB. Pipeline auto-triggers on every MDX update. The pilot produced 4,300 AI prototypes in 4 months using React design system components and Indeed Visual Language.
  • New York State (Jesse Gardner): Web components built with Lit and TypeScript, Figma parity via Code Connect. Custom MCP server exposes components and tokens documented via JSDoc with usage guidance baked in. Rich JSDoc → custom element manifest → MCP server. Fed a 5-page foster-adoptive parent PDF into Claude Code to generate a working multi-step NY State-styled form in 13 minutes.
  • Figma (Laura Fehre): Experiment translating Figma Make output (React, Tailwind, shadcn, Radix) into SwiftUI for a Notes app clone. Cursor parsed Apple Human Interface Guidelines URL into structured markdown files, drag-and-dropped into Figma Make's code editor with a mapping file translating Radix components to HIG components.
  • The Design System Guide (Romina Kavcic): Tidy Figma plugin with 66 MCP tools: audits naming, scores health across 6 categories, validates new variables, composes patterns (login form, destructive confirmation) from existing components. Paired with Observatory, a dashboard visualizing the knowledge graph across Figma, GitHub, Storybook, Linear, Chromatic, Playwright and PostHog.

How to build your design system MCP

  1. Start small (Romina Kavcic): Just add and document your naming conventions. Then add your 10 components or maybe 5, it's enough. Plant seeds, not trees.
  2. Codify foundations first (Romina Kavcic): Naming conventions, token taxonomy, component intent. Structure before automation.
  3. Pick a parser strategy (Diana Wolosin, Indeed): Diana built JavaScript parsers per knowledge domain (a11y, dev, localization, design) that read MDX and emit JSON. Cursor wrote them iteratively to handle every edge case.
  4. Use JSON for structured data (Diana Wolosin, Indeed): JSON for component APIs, props, sizes, variants. Markdown with frontmatter for natural-language rules. Don't dump MDX as-is.
  5. Build the pipeline (Diana Wolosin, Indeed): Ingestion → chunking (as sharp as possible) → indexing → vector database. Vectra is open source and a solid starting point.
  6. Automate freshness (Diana Wolosin, Indeed): Trigger the parser and ingestion pipeline on every MDX update so metadata is always current. Indeed's pipeline does this automatically.
  7. Add always-on rules and AGENTS.md (Diana Wolosin, Indeed): Foundations (typography, spacing, color, icon conventions) go into always-on rules so the LLM can't escape them. MCP is on-demand, rules are safety net.
  8. Benchmark before committing (Diana Wolosin, Indeed): Don't guess. Test multiple formats with the same prompts. Measure cost, accuracy and hallucinations. Diana spent 3 months on this and it paid off.
  9. Add usage guidance, not just APIs (Jesse Gardner, New York State): JSDoc with 'use filled for primary action, outline for secondary.' Guidance is useful for both humans and AI.
The Complete Guide

Design Systems MCP

What a design system MCP is, why it matters and how Diana Wolosin's Indeed benchmark proved JSON beats Markdown by 5x.

Based on real talks from 4 speakers at the AI Design Systems Conference 2026.

Definition

What is a design system MCP?

A Model Context Protocol (MCP) is a standardized way to give an AI agent on-demand access to external tools and structured knowledge. A design system MCP codifies your components, tokens, foundations and usage guidance into structured machine-readable metadata that lives in a vector database, so LLMs can query it and generate UI code that conforms to your system instead of guessing the average of the internet.

Want the full story on design system MCPs?

Diana Wolosin walks through her complete benchmark methodology, all 8 MCP configurations and the full JSON vs Markdown findings in her conference talk.

Get all recordings

The numbers

Real stats from Diana Wolosin's MCP benchmark at Indeed and other speakers.

$300 vs $1,500

per year: JSON vs Markdown MDX for the same 22-query workload

Source: Diana Wolosin, Indeed

5x

cheaper: structured JSON beats Markdown on the exact same design system

Source: Diana Wolosin, Indeed

80%

fewer tokens with JSON at equal or better accuracy vs hybrid Markdown

Source: Diana Wolosin, Indeed

1,056

prompts tested across 8 MCP configurations in the benchmark

Source: Diana Wolosin, Indeed

4,300

AI prototypes generated at Indeed in 4 months using their design system MCP

Source: Diana Wolosin, Indeed

50-80k

tokens wasted when you point AI at raw custom elements code

Source: Jesse Gardner, New York State

Diana's MCP benchmark

The definitive study on JSON vs Markdown for design systems MCPs. 1,056 prompts, 3 months, real Indeed design system.

Setup

Cursor plus Claude Sonnet 4.5 plus Indeed's design system MCP, stored in Vectra (open source vector DB). 22 prompts run 3 times each across 8 MCP configurations for a total of 1,056 prompts.

Formats tested

Markdown MDX (the original human docs), plain Markdown, hybrid Markdown + JSON, JSON (baseline and variations), and TOON (Token-Oriented Object Notation designed for AI).

Metrics

Input quality (cost, average tokens, relevance, query coverage) and output quality (coverage, hallucinations, average props found).

Finding #1

MCP is deterministic, LLM is stochastic. Same input to MCP returns the same output. Same MCP results sent to the LLM produce different reasoning each time.

Finding #2

Structured data mitigates AI uncertainty. Hybrid (MD + JSON) showed the most variation across runs and was unreliable. JSON showed the most consistency.

Finding #3

JSON has the highest accuracy when verified against actual documentation, and uses 80% fewer tokens than hybrid Markdown for equal or better accuracy.

Winner

JSON. Sharpest responses at the lowest token cost. Diana built this benchmark in ~3 months, vibe-coded with Cursor.

Key concepts

AI is a new user

Your design system was written for humans. The new user is AI, and it needs structured metadata, not documentation prose.

Diana Wolosin, Indeed

MCP deterministic, LLM stochastic

Same MCP query returns the same data every time. Same MCP results sent to the LLM produce different reasoning each time. Structured data mitigates that randomness.

Diana Wolosin, Indeed

JSON for structured data, Markdown for rules

JSON for MCP component APIs (props, sizes, variants): it's a contract. Markdown for natural-language instructions and rules for the LLM.

Diana Wolosin, Indeed

Progressive disclosure of context

Always-on rules for foundations (spacing, color, typography), MCP on-demand for components, AGENTS.md as orchestration layer. Together they form a plugin.

Diana Wolosin, Indeed

Foundations must be always-on

MCP only returns what you asked for. If the prompt says 'build me a card,' the LLM fills gaps with assumptions. Spacing and color belong in always-on rules.

Diana Wolosin, Indeed

Guidelines are not laws

Writing more rules into a guideline file doesn't control the outcome. In nearly 100% of cases the prompt will win over the guidelines.

Laura Fehre, Figma

Don't just plug MDX into an MCP

Connecting human documentation to an MCP is 5x more expensive and less accurate than structured JSON. Diana's benchmark proved this on Indeed's real design system.

Diana Wolosin, Indeed

Code as source of truth for MCP

JSDoc on Lit web components feeds the MCP with props, types, and usage guidance. Design and code stay in sync through a single structured source.

Jesse Gardner, New York State

Recordings available now

Diana's 1,056-prompt benchmark explained in full

Diana Wolosin spent 3 months building and running this benchmark on Indeed's real design system. She covers the full setup, every format tested and what to do next in her talk.

From the speakers

Real implementations

How Indeed, New York State, Figma and The Design System Guide are actually shipping design systems MCPs today.

IndeedDiana Wolosin

77 components processed via JavaScript parsers (one per knowledge domain: accessibility, development, localization, design) generating structured JSON. Original MDX in GitLab → parsers built with Cursor → JSON → ingestion into Vectra vector DB. Pipeline auto-triggers on every MDX update. The pilot produced 4,300 AI prototypes in 4 months using React design system components and Indeed Visual Language.

About this session →
New York StateJesse Gardner

Web components built with Lit and TypeScript, Figma parity via Code Connect. Custom MCP server exposes components and tokens documented via JSDoc with usage guidance baked in. Rich JSDoc → custom element manifest → MCP server. Fed a 5-page foster-adoptive parent PDF into Claude Code to generate a working multi-step NY State-styled form in 13 minutes.

About this session →
FigmaLaura Fehre

Experiment translating Figma Make output (React, Tailwind, shadcn, Radix) into SwiftUI for a Notes app clone. Cursor parsed Apple Human Interface Guidelines URL into structured markdown files, drag-and-dropped into Figma Make's code editor with a mapping file translating Radix components to HIG components.

About this session →
The Design System GuideRomina Kavcic

Tidy Figma plugin with 66 MCP tools: audits naming, scores health across 6 categories, validates new variables, composes patterns (login form, destructive confirmation) from existing components. Paired with Observatory, a dashboard visualizing the knowledge graph across Figma, GitHub, Storybook, Linear, Chromatic, Playwright and PostHog.

About this session →

Full case studies in the recordings

Indeed's 77-component pipeline, New York State's JSDoc-to-MCP architecture and Figma's cross-platform translation experiment are all covered in detail in the full talks.

Get all recordings

How to build your design system MCP

Nine practical steps synthesized from Diana, Jesse and Romina.

01

Start small

Just add and document your naming conventions. Then add your 10 components or maybe 5, it's enough. Plant seeds, not trees.

Romina Kavcic

02

Codify foundations first

Naming conventions, token taxonomy, component intent. Structure before automation.

Romina Kavcic

03

Pick a parser strategy

Diana built JavaScript parsers per knowledge domain (a11y, dev, localization, design) that read MDX and emit JSON. Cursor wrote them iteratively to handle every edge case.

Diana Wolosin, Indeed

04

Use JSON for structured data

JSON for component APIs, props, sizes, variants. Markdown with frontmatter for natural-language rules. Don't dump MDX as-is.

Diana Wolosin, Indeed

05

Build the pipeline

Ingestion → chunking (as sharp as possible) → indexing → vector database. Vectra is open source and a solid starting point.

Diana Wolosin, Indeed

06

Automate freshness

Trigger the parser and ingestion pipeline on every MDX update so metadata is always current. Indeed's pipeline does this automatically.

Diana Wolosin, Indeed

07

Add always-on rules and AGENTS.md

Foundations (typography, spacing, color, icon conventions) go into always-on rules so the LLM can't escape them. MCP is on-demand, rules are safety net.

Diana Wolosin, Indeed

08

Benchmark before committing

Don't guess. Test multiple formats with the same prompts. Measure cost, accuracy and hallucinations. Diana spent 3 months on this and it paid off.

Diana Wolosin, Indeed

09

Add usage guidance, not just APIs

JSDoc with 'use filled for primary action, outline for secondary.' Guidance is useful for both humans and AI.

Jesse Gardner, New York State

Common mistakes

Warnings from speakers who shipped and learned.

Don't just plug MDX into an MCP

It's 5x more expensive and less accurate than JSON. Diana's benchmark proved this on Indeed's real system.

Don't expect rules alone to control output

Laura: in nearly 100% of cases the prompt will win over the guidelines. You need structured data, not more rules.

Don't put everything in one big markdown file

Laura: single monolith overstretches the context window and breaks tasks. Split into multiple smaller files.

Don't rely on MCP for foundations

MCP is on-demand. Spacing, color and typography must be in always-on rules or the LLM will invent its own.

Don't point AI at the raw codebase

Jesse: 50-80k tokens fills the context window. Use MCP to expose only what's needed.

Don't skip designers in the loop

Jesse: vibe coding without systems thinkers recreates the silos we're trying to break down.

Frequently asked questions

Related guide

Agentic Design Systems: The Complete Guide

MCP is one piece of the puzzle. Learn how teams at GitHub, Indeed and New York State are building full agentic design systems with self-healing loops, trust levels and agent-ready architectures.

Watch the full talks

Everything on this page comes from 4 of 15+ talks at the AI Design Systems Conference 2026. Get lifetime access to all recordings, slides, templates and prompts.