Lead Design System · 2026 · Design System, MCP, AI Design Ops

Carrefour — Marcel Design MCP

carrefour.design

The design system already had the answers Before: Figma specs, tech specs, carrefour.design, legacy files and design debt float apart, unreadable by AI. Now: specs, tokens, UX writing, patterns and the glossary orbit Marcel MCP, readable by agents. A new channel, not new content. BEFORE · SCATTERED Figma specs Tech specs carrefour.design Legacy Design debt NOW · CENTRED ON MARCEL Marcel MCP specs tokens UX writing patterns glossary

Marcel, Carrefour's design system, had two ways in: designers opened the Figma library, developers installed the package. AI tools had none, so they invented. I gave Marcel a third way in — an API that any AI assistant queries before it draws — and the same rules now reach everyone who builds: designers, developers and product teams.

1
Pass, not 3–4
÷2
Figma Make credits
35
Tools
3
Audiences

01 — AI, the third user

By 2026, vibe design was everywhere. Figma Make, Cursor, Lovable, v0: anyone could generate a plausible screen in minutes. Working alongside Carrefour product teams, I saw the other side of it — fast, plausible and off-brand. A random blue, an 8px radius, a "Submit" button, a table full of "John Smith".

A human fills the gaps in a design system with taste. A model fills them with invention. Every invention is a correction, every correction a paid round trip, and user tests end up reacting to the mockup instead of the journey.

Guess, or query the source

Guessing versus querying the design system Without a source of truth the model answers from memory, ships a plausible but off-brand screen and loops through three to four paid corrections; with Marcel MCP it queries exact values and ships an on-brand screen in one pass. WITHOUT A SOURCE OF TRUTH GUESS RE-PROMPT × 3–4 IN Prompt natural language LLM Model answers from memory OUT Plausible screen random blue · 8px radius FIX Manual correction a paid round trip WITH MARCEL MCP ONE PASS QUERY EXACT VALUES IN Prompt @Marcel … LLM Model asks, never guesses MCP Marcel MCP source of truth OUT On-brand screen tokens · specs · wording LEGEND Designer input Source of truth Wasted step Loop / return
Without a source of truth, the correction loop stays with the designer. With one, the model asks before it draws.

02 — What API-izing a design system means

An API is a contract: ask a precise question, get a precise answer, in a format any program can read. API-izing Marcel did not change its content. It changed how you reach it. The designer reads it, the developer installs it, the machine queries it — through the same door, under the same rules.

The Model Context Protocol is the standard plug between an AI and a source of truth. Plug the design system into the standard rather than into one tool, and every assistant that speaks MCP speaks Marcel.

Same door, three users

Three ways into the same design system The designer reads and interprets the Marcel library, the developer installs and imports the package, and the AI assistant queries the server every time; all three draw on the same rules, tokens, components, patterns, words and accessibility to build a screen that looks like the product. READ & INTERPRET INSTALL & IMPORT ASKS, EVERY TIME BUILDS Designer opens the library Developer installs the package AI assistant queries the server MARCEL rules & tokens components & icons patterns & templates dataviz & motion words & data accessibility one source of truth, for everyone On-brand screen looks like the product same door, same rules
The thick line is the whole point: where a person fills the gaps with years of craft, a model needs everything written down.

The contract, concretely

The AI asks, Marcel answers Examples of the API contract: the AI asks which components exist, the exact specs of a button, Modal or Drawer, how to word an error and what data fills a table; each question maps to one Marcel MCP tool that returns a precise answer such as radius 4px, 56px and #0970e6. THE AI ASKS IT CALLS A TOOL MARCEL ANSWERS “Which components exist?” get_components 69 components, by family “Exact specs of a Button?” get_component_specs radius 4px · 56px · #0970e6 “Modal or Drawer?” get_decision_framework a rule, with its exceptions “How do we word this error?” get_content_library microcopy validated in prod “What data fills a table?” get_sample_data 14 Carrefour stores, seeded
Each question the model would otherwise answer from memory maps to a tool that returns the real value.

An LLM that invents is guessing. Marcel knows.

What the API serves

What Marcel knows, layer by layer Five layers of knowledge served by Marcel MCP, from foundations (tokens, typography, icons) through components, patterns and decisions, voice and content, up to realistic data; voice and content is highlighted as the layer most AI tools lack. L5 Realistic data get_sample_data · get_dataviz_* 13 domains · 14 charts L4 Voice & content get_ux_writing_rules · get_glossary · get_content_library 5C · 80 terms · 49 copies L3 Patterns & decisions get_page_patterns · get_decision_framework 8 patterns · 9 frameworks L2 Components get_component_specs · get_component_variants 69 components · 330 variants L1 Foundations get_tokens · get_typography · get_icons 105 tokens · 500 icons SCREEN SYSTEM
Form is not enough: the right component says the wrong thing if it speaks badly. Voice and content is the layer generic AI tools never have.

Who it serves

The circle of people the design system serves keeps widening Marcel was first consumed by hand by designers as a library, then in code by developers as a package, and now through the MCP by product, data and marketing teams as a service they query. CONSUMED BY HAND Designers a library you open CONSUMED IN CODE + developers a package you install CONSUMED VIA MCP + product, data, marketing a service you query
The same investment, served to more people each time. The MCP doesn't add a channel, it unifies them.

03 — For designers

Prototypes that look like the product, from the first prompt. In Figma Make there is nothing to install: Marcel is deployed as a connector for the whole organisation. Type @Marcel in the chat and the model picks among 35 tools on its own — the designer never drives a tool.

What changes: one pass instead of three or four corrective iterations, user tests that test the journey again, half of all revisions sent back to the canvas at zero credits, and the whole system — catalogue, tokens, icons — explorable without opening the nine Figma libraries.

One sentence, orchestrated

One prompt, orchestrated by build_screen A designer types one sentence in Figma Make; the build_screen prompt returns a locked work plan, then the model calls get_page_patterns, get_component_specs and get_ux_writing_rules on Marcel MCP before delivering an on-brand screen on the first pass. @Marcel drive order tracking prompts/get build_screen locked work plan plan + 35 tools get_page_patterns structure · mandatory states get_component_specs × n 4px · 56px · #0970e6 get_ux_writing_rules 5C · glossary · microcopy on-brand screen, first pass IN Designer one sentence HOST Figma Make connector, no setup LLM Model picks the tools MCP Marcel MCP answers with facts LEGEND Source of truth Call Return Delivered screen
build_screen turns one sentence into a locked plan, and the model fetches structure, exact specs and wording before drawing anything.
Figma Make generating a full Carrefour screen from a single @Marcel prompt, calling get_component_specs and get_design_rules
suggest_next_steps returning a compliance audit of the order tracking screen in Figma Make

build_screen in production: "@Marcel I want a drive order tracking screen, mobile, light mode" — Marcel components, resolved tokens and approved copy, on the first pass.

Then one question — "what is missing from this order tracking screen?" — and the model calls get_components, get_ux_writing_rules, get_page_patterns and get_glossary on its own, and returns a compliance audit: components to swap, wording to fix, critical gaps.

Order tracking screen before the audit: custom cards and generic wording
Order tracking screen after the audit: Functional Notice, driver Avatar, product Accordion and rating, all Marcel

Same screen, before and after the audit: a Functional Notice in the Info tone, a driver section with Avatar and call buttons, an Accordion of products with substitutions and totals, and wording that follows Marcel's Constructive principle — "Votre commande arrive dans environ 45 min." Two prompts, zero specs written by hand.

An assistant that says "do it yourself"

revise_design: the credit guard Before any tool call, revise_design classifies a change request: small edits such as a few words, a spacing under 8px or a token swap stop with instructions to do it by hand at zero credits; larger changes become a surgical edit that is restated before being built. YES NO Change request on an existing screen Worth a prompt? PROMPT Surgical edit only the target moves Restated, then built “I change X, I keep …” STOP Do it by hand 0 tools · 0 credits ≤ 5 words of copy, spacing under 8px, a token swap, a boolean to flip.
Most AI tools encourage you to re-prompt, and every turn is billed. Marcel classifies the request first, and half the time sends you back to the canvas.
The Marcel component catalogue rendered inside a Claude conversation as an MCP App: 69 components by family
The Marcel token explorer inside a Claude conversation, dark mode, 105 semantic tokens with their hex values

Marcel also answers with interfaces, not paragraphs: the component catalogue, the token explorer and the 501 icons open inside the conversation as MCP Apps. Search, filter, click — without leaving the chat.

04 — For developers

The same facts, compiled to your stack. Clone the repository and the server runs in Claude Code, Cursor, Cline, Windsurf or Zed. The coding assistant stops guessing paddings: it asks get_component_specs and gets 4px, 56px, #0970e6 — resolved values, not token names a framework would ignore.

For tables, get_table_recipe ships MarcelDataTable: React on TanStack Table v9 — the library on Carrefour's TechRadar — with CSS variables generated from the tokens. What Marcel generates rests on the same foundation as what the developers build.

Facts, not code

One neutral token, compiled by each client The semantic token Action/Background/Brand, resolved to #0970e6, is served as neutral data and compiled by each client to CSS, SwiftUI or Flutter. COMPILED BY THE CLIENT TOKEN Action/Background/Brand #0970e6 · light & dark WEB CSS --brand: #0970e6; IOS SwiftUI Color(hex: 0x0970e6) APP Flutter Color(0xFF0970E6)
The server serves neutral data and never presumes the stack. The client decides what a token becomes.

One server, every assistant

One server between the sources and every assistant Figma libraries, Zeroheight documentation, team standards and the MCP specification are aggregated into the Marcel MCP server, which serves designers in Figma Make, developers in Claude Code and Cursor, and product teams in Lovable and ChatGPT. SOURCES AUDIENCES Figma libraries 9 libraries · 69 components Zeroheight docs exported to markdown Team standards rules · conventions MCP specification the protocol itself MCP Marcel MCP neutral JSON · Vercel built with Claude Code Designers Figma Make · @Marcel Developers Claude Code · Cursor Product teams Lovable · ChatGPT
Plugged into a standard, not a tool: the next AI assistant ships already speaking Marcel, with no integration project — and the choice of model stays reversible.
01

Scraped from Figma, not from documentation

Specs come from the components themselves in the 9 Marcel libraries. The first scrape already corrected a belief: Marcel's default radius is 4px, not 8px.

02

Static JSON, zero runtime calls

Data is bundled with a stateless server on Vercel and read in milliseconds: no live Figma call, no rate limit, no API token in production. Metrics are written after the response, so monitoring adds no latency.

03

Kept in sync

A drift check runs in CI against the Figma REST API, and a sync skill re-extracts what changed while preserving 249 hand-written passages.

04

Next: a design gate in CI

Today the specs plate Marcel onto any stack to prototype. Tomorrow the same specs drive Carrefour's own web components and lint every front-end pull request before merge.

05 — For product teams

Anyone who can write a sentence. A product manager mocks up a need without waiting for a designer, and the screen already follows the rules. clarify_design_brief asks the right questions in a single card — goal, key elements, constraints, definition of done — and the decision frameworks settle Modal-or-Drawer without a review meeting.

Stakeholders react to content that looks real: get_sample_data fills screens with credible, seeded Carrefour data — store formats, real house brands, delivery slots — and zero real data. And for teams building chat agents, Marcel themes OpenAI ChatKit widgets from its own tokens and audits them.

The right questions, before the first pixel

build_screen: three locks before the first pixel A brief is checked for completeness; if fields are missing, clarify_design_brief asks every question in one card and the answers complete the brief; once complete, mandatory tool calls and critical rules run before the screen is generated. YES NO ANSWERS Brief goal · device · mode Brief complete? STEP 1 Mandatory calls specs, tokens, copy STEP 2 Critical rules 4px · Ubuntu · 5C Screen on-brand STEP 0 One question card clarify_design_brief
An incomplete brief costs three to four iterations. Step 0 refuses to generate until a single question card has collected everything.

Same rules, without the brand

Two surfaces from one server The same Marcel MCP server exposes a branded surface for designers in Figma Make, with the #0970e6 brand CTA, and an unbranded wireframe surface for product managers in Lovable, where six hex substitutions turn the brand into grey while structure, components and rules stay identical. FULL BRAND 6 HEX SWAPS MCP Marcel MCP one server HI-FI Branded surface /mcp · Figma Make WIRE Unbranded surface /mcp/wireframe Lovable Designers CTA #0970e6 a finished-looking screen Product managers CTA #696969 same structure & rules
PMs on Lovable get the same Marcel — components, radius, icons, typography, writing, data — without the brand colour. A screen a designer can pick up downstream, that never passes for a finished deliverable. Feedback colours stay, so an error still reads as one.

06 — Measured, shipped

AI costs money, and framing is what limits the investment. Three levers cut calls without losing quality: a complete brief from the first prompt, rules attached once per session as memory, and gradient deduplication — a repeated call returns the full answer, then a reminder, then a simple reference. Efficiency is observed, not assumed: a dashboard tracks calls, latency, errors and who calls what.

Marcel MCP monitoring dashboard overview: tool calls, health, top tools and tool × client matrix (demo data)
Marcel MCP monitoring detail: usage heatmap, call sequences and archived prompts (demo data)

The teams' first reflex was exact values, not exploration: get_component_specs leads production usage, ahead of get_design_rules and the system overview. First-prompt compliance halves Figma Make credits, with a 0% tool error rate.

March → September 2026

Shipped in iterations, March to September 2026 Release timeline of Marcel MCP in 2026: first tools on 20 March, credit efficiency on 28 May, realistic data on 11 June, the catalogue in chat on 29 July, 501 icons on 31 July, the unbranded wireframe surface on 6 August, data tables on 16 September and ChatKit widgets on 22 September. MAR APR MAY JUN JUL AUG SEP First tools v1.0 · 5 tools Credit efficiency v1.5 · memory · brief card Realistic data v1.6 · get_sample_data Catalogue in chat v1.7 · MCP Apps 501 icons v1.8 · real glyphs Wireframe surface PMs on Lovable Data tables v1.9 · TanStack Table ChatKit widgets ChatGPT surfaces
242 commits in six months. Each release is named after the problem it removes, not the feature it adds.

07 — What's next

Reading is only half the way. Today Marcel answers "the radius is 4px". Next it generates the compliant component, detects the drift and pushes the fix — audits of a franchisee's live site, a design gate on every pull request, accessibility checked continuously ahead of RGAA 2026, and onboarding where the design system answers for itself. No more evangelism: a design system plugged into the pipeline proves its value by running, not by explaining.

Design System to Design Intelligence Platform: Specs, Explore, Production, QA and Doc agents orbit the platform, plugged into an API layer (deploy, lint, generate, orchestrate) and a context layer (rules, patterns, design intent, tokens)
From a read API to an operating layer — generate(), lint(), deploy(), orchestrate() — with agents orbiting the design system the way components once orbited the library.

The design system is no longer documentation you read. It is a service you plug in.