v0.1.0 — now available on npm

Give your AI assistant the power to design.

A lightweight MCP server that lets Claude Code, Codex, Cursor, VS Code/Copilot, and Windsurf generate, refine, and adopt production-ready UI designs.

Read the Docs
Built for
Claude Code, Codex, Cursor, VS Code, Windsurf
Codex and Windsurf use user-scope setup today

Design as a compilation target.

The MCP server treats UI generation as a structured process — not simple text completion. It handles layout, styling, and preview rendering so your agent can iterate visually.

Production-grade design

Fine-tuned models generate clean, professional UI — not generic templates. Tailwind CSS output with proper hierarchy and real typography.

Iterative refinement

Refine any design with natural language feedback. Adjust layouts, swap color schemes, or change spacing — without starting over.

Repo-aware context

Detects your framework, component library, CSS tokens, and routes. Designs are generated to fit your existing stack and conventions.

Zero-friction setup.

One command registers the MCP server with your coding assistant. OAuth opens your browser once for sign-in, and the CLI writes the right config format for Claude Code, Codex, Cursor, VS Code, or Windsurf.

.cursor/mcp.json
{
  "mcpServers": {
    "aidesigner": {
      "url": "https://api.aidesigner.ai/api/v1/mcp"
    }
  }
}

Twenty-one tools. One protocol.

The MCP server exposes a set of tools your AI coding assistant can call directly — design generation, live canvas streaming, and account info.

generate_design

Create a complete UI design from a text prompt. Supports desktop and mobile viewports with prompt-driven generation.

refine_design

Iterate on a previous design with natural language feedback. Adjusts layout, colors, typography, or content without starting over.

generate_image

Generate or edit standalone imagery. Supports reference images, saved brand kits, and paired-canvas delivery.

generate_website_design_image

Create premium website reference images and landing page slices, optionally steered by a saved brand kit.

generate_branding_kit_variations

Generate a square 3x3 board of nine distinct brand identity directions from a brand brief.

generate_media_asset_kit

Generate a square 3x3 reusable media asset kit and, when paired to the editor, auto-extract nine transparent asset canvases.

create_brand_kit_from_variation

Save a reusable brand kit from the selected tile in a 3x3 identity exploration board.

list_brand_kits

List saved brand kits with public summaries, thumbnails, and swatches so the agent can reuse an existing direction.

set_editor_brand_kit

Set or clear the active saved brand kit for the paired editor project.

save_media_asset_to_brand_kit

Save an existing image canvas as a named brand kit media asset with intended-use guidance for future designs.

drop_image_on_canvas

Place an existing image file or safe image URL onto the paired editor canvas without regenerating it. PNG transparency is preserved.

remove_image_background

Remove the background from an existing image canvas as a Pro perk. Creates a new transparent PNG canvas beside the original.

vectorize_image

Convert an existing raster image canvas to SVG as a Pro perk. Creates a new vector image canvas beside the original.

extract_image_assets

Extract up to 12 embedded photos, logos, illustrations, textures, text-bearing graphics, or media shots from an image canvas into reusable image canvases.

create_editor_session

Spin up a new editor project and auto-pair the agent to it. Returns a URL you open to watch designs stream in live.

link_editor_session

Pair the agent to an existing project using a 6-character code from the editor. Designs then deliver straight to the canvas.

unlink_editor_session

End the pairing between the agent and the editor session.

list_canvases

List every design in the paired editor session, newest-first, so the agent can see what's already there.

get_canvas

Fetch the full HTML and metadata for a single canvas by id, after finding it with list_canvases.

extract_canvas_design

Resolve a natural-language query like "the pricing page" to the matching canvas via fuzzy search across names and content.

get_credit_status

Check your remaining credits, monthly usage, and subscription status. Free tier included.

whoami

Returns your connected AIDesigner account identity and authorized scopes.

URL-based workflows

Clone, enhance, or get inspired by any website.

Pass a reference URL to the MCP server alongside your prompt. AIDesigner analyzes the page and uses it as a starting point — whether you want an exact clone, a modernized version, or just the visual style.

Clone

Near-1:1 recreation of any website. Paste a URL and AIDesigner reproduces the layout, typography, colors, and content — ready to edit and make your own.

"Clone this landing page: https://example.com"

Enhance

Modernize an existing site while keeping its content. AIDesigner analyzes the URL and rebuilds it with better spacing, hierarchy, and visual polish.

"Enhance this page with a more modern design"

Inspire

Use any URL as a style reference without copying its content. AIDesigner extracts the visual language — colors, layout patterns, typography — and applies it to your prompt.

"Design a pricing page inspired by this site's style"

Try the browser-based website cloner →Or use clone mode directly from your coding assistant via MCP
Our Community Loves Us

What people are saying

Frequently asked questions

MCP (Model Context Protocol) is an open standard that lets AI coding assistants connect to external tools. The AIDesigner MCP server gives these assistants the ability to generate and refine UI designs directly within your development workflow.

AIDesigner currently supports Claude Code, Codex, Cursor, VS Code/Copilot, and Windsurf. Codex and Windsurf are user-scope installs today, and VS Code requires GitHub Copilot in Agent mode.

The MCP server uses OAuth. When you first connect, your MCP client opens a browser window for sign-in. After that, tokens are managed automatically — no API keys to configure.

The MCP server generates complete HTML documents with inline Tailwind CSS. The CLI then renders PNG previews and generates adoption briefs that map the design to your framework (Next.js, React, Vue, etc.).

Free accounts include credits to try the tool. Premium plans offer higher limits. You can check your balance anytime with the get_credit_status MCP tool.

Yes. The server analyzes your repo for CSS tokens, component libraries (Radix, shadcn/ui), and styling systems (Tailwind, CSS variables). Generated designs reference your existing tokens and patterns.

Start designing from your terminal

One command to connect. Zero context switching. Production-ready UI in seconds.

Get Started Free

Used by developers building with Claude Code, Codex, Cursor, and VS Code

AI UI Design MCP Server | AIDesigner | AI Designer