Atelier-MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Atelier-MCPinspect my project and prepare a design plan to modernize the landing page"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
The problem
You ask an AI to "make the hero section look better" and it dumps 200 lines of arbitrary CSS, overwrites your design tokens, ignores your existing component library, and produces something that looks like every other AI-generated site.
Atelier fixes that.
Related MCP server: ui-architect-mcp
What this actually does
Atelier MCP sits between your AI coding assistant and your frontend project as an MCP server. It exposes 10 tools that give the AI structured context about your project before it touches anything:
Your AI assistant
│
▼
┌──────────────┐ inspect → understand tokens, components, conventions
│ Atelier MCP │ search → find recipes that match the project's aesthetic
│ (stdio) │ plan → structured changes with rollback capability
└──────────────┘ apply → safe changes with automatic backups
│
▼
Your website projectThe AI inspects first, consults a curated recipe catalog, builds a plan, and only then makes changes — with backups so you can undo everything.
One-line install
macOS / Linux:
curl -fsSL https://raw.githubusercontent.com/YOUR_USERNAME/atelier-mcp/main/install.sh | bashWindows (PowerShell):
irm https://raw.githubusercontent.com/YOUR_USERNAME/atelier-mcp/main/install.ps1 | iexBoth scripts check prerequisites (Node 20.19+ (20.x) or 22.12+, Git), clone the repo, install
dependencies, build, and drop the atelier CLI onto your PATH. Done.
Quickstart
# clone it
git clone https://github.com/YOUR_USERNAME/UiMCP.git
cd UiMCP
# install
npm install
# build everything
npm run build
# install into your website project
npx tsx packages/cli/src/cli.ts install --source . /path/to/your/site
# or just run the MCP server directly for debugging
npm run dev:serverHook it up to Claude Code
claude mcp add atelier-mcp -- npx tsx /path/to/atelier-mcp/packages/server/src/index.tsThat's it. Your AI now has design tools.
Monorepo layout
.
├── packages/
│ ├── server/ → the MCP server itself (10 tools over stdio)
│ ├── cli/ → install/uninstall/doctor/update commands
│ ├── recipes/ → curated design & motion recipe catalog
│ ├── adapters/ → framework detection (React+Vite, Next.js)
│ └── monitor/ → Neurolink web dashboard (real-time visualizer)
│
├── install.sh → one-line installer (macOS/Linux)
├── install.ps1 → one-line installer (Windows)
├── LICENSE → MIT
├── THIRD_PARTY_LICENSES.md
└── package.json → npm workspaces monorepo rootThe tools
Tool | What it does |
| Reads your project — framework, components, design tokens, config files, existing conventions. The AI calls this first. |
| Searches a curated catalog of motion & UI patterns by purpose, aesthetic, framework, or performance cost. |
| Returns full implementation code for a recipe — React component, CSS, dependencies, a11y metadata, license info. |
| Creates a structured plan: design direction, tokens, selected recipes, and exact file changes. Persisted to disk. |
| Shows what would change before anything is written. Detects conflicts, lists dependencies. |
| Writes the files. Backs up everything it touches. Returns a record ID for rollback. |
| Reverts Atelier changes. Detects if you've edited files since and warns before overwriting. |
| Screenshots your running site at mobile/tablet/desktop via Playwright. |
| Runs accessibility checks (axe-core), Lighthouse performance, and contrast validation. |
| Event log and session state — what tools were called, what changed, project dependency graph. |
Recipe catalog
The recipe catalog isn't just "here's some CSS." Each recipe ships with:
Full React component code for both Vite and Next.js
Accessibility story — keyboard, touch, screen reader,
prefers-reduced-motionPerformance budget — JS size, CSS size, GPU needs, expected FPS
License — know exactly what you're shipping
Parameters — every recipe is configurable, not copy-paste-and-pray
Current categories:
Category | Examples |
Text reveals | Split text, per-word stagger, line-by-line |
Entrances | Staggered grid cascade, fade-scale-in |
Buttons | Tactile press feedback, ripple effects |
Hover effects | Perspective tilt card, magnetic cursor |
Dialogs & menus | Accessible modal with focus trap, slide panels |
Route transitions | Cross-fade, slide, scale between pages |
Scroll progress | Reading progress bar |
Parallax | Restrained depth layers (not the 2014 kind) |
Sticky storytelling | Scroll-pinned narrative sections |
CLI
atelier install [dir] # Install Atelier into a project
atelier uninstall [dir] # Clean removal
atelier doctor [dir] # Health check
atelier update [dir] # Update source reference
atelier serve # Run MCP server directlyInstall flags:
--source <path> Local Atelier source directory
--repo <url> Clone from a git repo instead
--ref <ref> Specific branch/tag/commit
--dry-run See what would happen
--skip-mcp Don't auto-configure MCP client
--skip-skill Don't install CLAUDE.md companion fileHow it stays safe
Never writes without a plan. The AI must call
prepare_design→preview_changes→apply_changes. No yolo file writes.Backs up everything it touches. Every modified file gets a backup in
.atelier/backups/.Rollback is first-class.
rollback_changesrestores backups and detects user edits made after Atelier's changes.Path validation. All file operations are sandboxed to the project directory. No traversal escapes.
Respects what's already there.
inspect_projectreads existing tokens, conventions, and component patterns so the AI doesn't bulldoze your design system.
Development
npm run typecheck # type checking across all packages
npm run lint # ESLint across source and tests
npm test # vitest
npm run dev:server # watch mode on the MCP serverSupported frameworks
React + Vite (≥18.0)
Next.js (≥14.0, App Router)
More coming. PRs welcome.
Neurolink Monitor
A responsive, dark dashboard for watching Atelier in action. Requires Node.js 20.19+ (20.x) or 22.12+ for the development toolchain.
# Start the monitor
npm run dev:monitor -- --project /path/to/your/site
# Open http://localhost:4510
# Use --port 4511 to choose a different portWhat you get:
Overview — Project metrics, an interactive tool map, recent activity, and plans.
Activity — Search and filter the latest 200 events; pause the feed to inspect details.
Project — Framework, dependencies, components, design tokens, and configuration files.
Design plans — Review saved plans, status, file changes, and related change records.
Tool library — Explore all ten tools, their purpose, and recent calls.
Snapshot export — Download the current workspace data as JSON.
The monitor connects locally, reconnects automatically, and displays errors when
data cannot be refreshed. Project-specific MCP events are read from the project's
.atelier/events/ directory. Use the same project path in your coding assistant
and monitor to see its activity. Press / to search and ? for the workflow guide.
Motion respects the system's reduced-motion preference.
For the optional MCP screenshot and browser audit tools, install Chromium once:
npx playwright install chromiumLicense
This project is dual-licensed. Pick whichever works for you:
MIT — LICENSE-MIT
Apache 2.0 — LICENSE-APACHE
Blue Oak 1.0.0 — LICENSE-BLUEOAK (covers bundled dependencies from
glob)
Third-party dependency licenses are fully documented in THIRD_PARTY_LICENSES.md.
This server cannot be deployed
Maintenance
Related MCP Connectors
Live React design-system APIs, patterns, and code validation so AI agents build real UI, not slop.
Serves your design system and coding standards to coding agents, so they stop guessing.
UI design from prompts, screenshots, and URLs for AI coding agents and theme tokens.
Access and maintain design system docs, tokens, components, skills, and contexts across any project.
Related MCP Servers
- FlicenseAqualityDmaintenanceProvides AI assistants with frontend development tools including component scaffolding, bundle analysis, accessibility checks, and responsive design guidance. Enables automated generation of React components with tests and stories, bundle optimization recommendations, and WCAG compliance fixes.6-
- FlicenseAqualityCmaintenanceEnables AI assistants to generate production-ready, professional UI design systems and components from simple descriptions, with real images, animated components, and automated quality checks.16-
- AlicenseNot gradedqualityBmaintenanceProvides AI-powered UI design, repo intelligence, and frontend code generation tools, enabling assistants to scaffold projects, generate and modify UI components, manage design tokens, and analyze repositories.MIT
- FlicenseNot gradedqualityBmaintenanceEnables AI coding agents to inspect a project's design tokens and component catalog, audit React/Tailwind files against seven static UI quality rules, score overall UI health, and apply auto-fixes that are re-audited and reverted on regression. It works both locally over filesystem access and remotely on raw TSX/JSX or CSS snippets passed directly as strings.-