Skip to main content
Glama

AI Canvas

Design systems your AI agent builds with and you reshape to your project, plus standalone React components and page blocks that fit any of them, installed as real source code in your project rather than as a dependency.

Install

Add any component to your project with one command:

npx shadcn@latest add @aicanvas/task-cards

Starting a new project? Initialize first, then add:

npx shadcn@latest init        # new projects only
npx shadcn@latest add @aicanvas/task-cards

Installs are tied to a free AI Canvas account. Signed out, the CLI writes a small placeholder file instead of the component, so authenticate once first:

  1. Sign up free and copy your token from account settings.

  2. Put it in .env.local as AICANVAS_TOKEN.

  3. Add the registry to your project's components.json, once:

{ "registries": { "@aicanvas": { "url": "https://aicanvas.me/r/{name}.json", "params": { "token": "${AICANVAS_TOKEN}" } } } }

Every npx shadcn@latest add @aicanvas/<slug> is authenticated after that. Each component page also shows a ready tokenized command when you are signed in.

No account is needed to read the code: every component page shows its full source, free to read and copy.

Requirements

Tailwind CSS v4, React 19 and Framer Motion. Components are written against Tailwind v4 tokens and will not render correctly on v3. The Next.js App Router is the default target, not a requirement.

Three ways to use it

Path

Command or action

Best for

shadcn CLI

npx shadcn@latest add @aicanvas/<slug>

Dropping finished, open-source code straight into your repo

AI Canvas MCP

npx -y @aicanvas/mcp

Letting your AI editor search and install components for you

Remix with AI

Copy the full prompt from any free component page

Rebuilding a component your way in any AI coding tool

Browse the full catalog and copy the exact command for any component at aicanvas.me. pnpm dlx, yarn dlx, and bunx work too.

Related MCP server: MCP Magic UI

Why AI Canvas

  • MIT licensed. The free library is MIT, so you can use it in personal and commercial projects, modify it freely, and ship it without attribution. Premium components, design systems, and templates are proprietary.

  • Full source, yours to keep. Every component arrives as real React and TypeScript code in your codebase, not a black-box dependency. Restyle it, extend it, or ship it as is. It is yours.

  • Built for AI workflows. Install with the shadcn CLI, connect the MCP so your agent installs for you, or hand it a remix prompt that works in any AI coding tool.

  • Animated by default. Built with Framer Motion and Tailwind CSS, ready for the Next.js App Router or any modern React setup. 3D pieces use Three.js.

Components

Blocks

Composed, multi-component page sections: card decks, task boards, galleries. Same install, same MIT license, just a larger unit of work than a single component.

Use it with your AI editor (MCP)

Connect AI Canvas to your AI editor and let your agent search, inspect, and install components for you. Save tokens. Do not start from scratch.

claude mcp add aicanvas -- npx -y @aicanvas/mcp

Or add it to your MCP config:

{
  "mcpServers": {
    "aicanvas": {
      "command": "npx",
      "args": ["-y", "@aicanvas/mcp"]
    }
  }
}

Works with Claude Code, Codex, Cursor, and other MCP-compatible editors. The server is read-only and fetches the live registry at runtime, so new components reach your agent shortly after they ship. It returns published component metadata and source.

Repository layout

This repo holds the AI Canvas source: the website and the free component library. The component source lives here, not just the site.

Path

What's there

components-workspace/<slug>/

Each free component: index.tsx source and its prompts.ts remix prompt, plus a spec.md on many

design-systems/andromeda/

The free Andromeda design system source: tokens, components, utilities

app/

The aicanvas.me website (Next.js App Router)

scripts/generate-registry.mjs

Builds the shadcn registry JSON from the sources above

The @aicanvas registry files served at /r/*.json are generated at build time from components-workspace/ and design-systems/, so there is no checked-in registry/ folder. Run node scripts/generate-registry.mjs to produce them locally.

Tech stack

React and TypeScript, Tailwind CSS, and Framer Motion. Built for the Next.js App Router and works in any modern React setup. 3D components use Three.js.

Common Questions

Do I need an account? Not to read the code. Every component page shows its full source, free to read and copy. Most of the free library also lives right here in this repo. A free account is only needed for the one-command npx shadcn add install, which writes a placeholder file when you are signed out.

Does it work outside Next.js? Yes. Components are plain React and TypeScript with Tailwind and Framer Motion. The App Router is the default target, not a requirement. A few components load their display font through next/font/google; outside Next.js, swap that import for your own font loader.

Do I get a dependency or real code? Real code. Installing copies the source into your project, so you own it and can change anything. There is no AI Canvas package to keep in sync.

Can I use it commercially? Yes, the free library is MIT. Use it in commercial projects, modify it, and ship it without attribution. Premium is licensed separately, see License.

What is the remix prompt? Every component ships with one comprehensive, platform-agnostic prompt. Paste it into any AI coding tool to rebuild that component your way instead of copying it as is.

How do I keep components up to date? You do not have to. Once installed, the code is yours and never changes under you. Re-run the install command if you want the latest version of a component.

Contributing

AI Canvas is actively maintained and new components ship regularly. Issues, ideas, and pull requests are welcome: open an issue to suggest a component or report a bug.

If AI Canvas saves you time, a star helps more builders find it.

License

The free library is MIT licensed. Use it in personal and commercial projects, modify it freely, and ship it without attribution. See LICENSE. Premium components, design systems, and templates are proprietary under the AI Canvas Premium License.

Available Tools

14 tools
get_audit_checklistGet the AI Canvas audit checklistA
Read-onlyIdempotent
Inspect

Return the checklist to run on a finished page or component built with AI Canvas parts: install completeness, prop correctness (via validate_usage), both themes, layout at 320px, motion cleanup and reduced motion, keyboard and contrast, and composition rules. Use it as the last step of a build, or whenever the user asks "is this done right".

Inputs: scope.

get_componentGet an AI Canvas component (full metadata + source code)A
Read-onlyIdempotent
Inspect

Return the complete record for a single component: metadata + the entire .tsx source code as a string, plus the AI Canvas icon files it imports. Use after the user picks one from a search/list result and wants to see the implementation, or before installing if they want to inspect first. The source is the exact file the CLI install would write into a project.

Inputs: slug (required).

get_component_propsGet the documented props (API) of an AI Canvas componentA
Read-onlyIdempotent
Inspect

Return the prop table of one component: every prop with its type, whether it is required, its default and what it does. Free, needs no account, and far smaller than the full source. Call it before writing JSX for a component you have not read, or to answer "what can I configure on X". A component with no table takes no props: it is self-contained and adapted by editing its source.

Inputs: slug (required).

get_icon_installGet the install command and import for an AI Canvas iconA
Read-onlyIdempotent
Inspect

Return the shadcn command that installs one AI Canvas icon as an editable file (components/icons/.tsx plus the shared IconBase runtime), with the exact import and JSX to use it. Line is free with an account; Duotone, Solid and Solid duotone are Premium. Find names with search_icons first.

Inputs: name (required), style.

get_install_commandGet the AI Canvas install command for a componentA
Read-onlyIdempotent
Inspect

Return the shadcn CLI command to install a single component into the user's project. After calling this, suggest the user run the command (or run it yourself if you have shell access). Requires shadcn-aware setup (Tailwind v4, components.json). For projects without that, fall back to get_component and write the file manually.

Inputs: slug (required).

get_systemGet a complete AI Canvas design system (every file)A
Read-onlyIdempotent
Inspect

Return all files for a design system in a single response (every component, plus shared tokens and utilities), ready to write into the user's project. Use when the user wants to adopt a whole system rather than pick individual components.

Inputs: slug (required).

get_templateGet a complete AI Canvas design-system template (every file)A
Read-onlyIdempotent
Inspect

Return all files for a single template: the example composition plus every component it uses plus shared tokens. Use for "I want exactly this dashboard" / "give me the entire mission-control screen". One CLI command installs all files.

Inputs: slug (required).

list_categoriesList AI Canvas categoriesA
Read-onlyIdempotent
Inspect

Return every category in the AI Canvas standalone component library, with the number of components in each. Use to orient before listing or searching, e.g. "what kinds of components are available?"

Inputs: no named parameters.

list_componentsList AI Canvas componentsA
Read-onlyIdempotent
Inspect

Return AI Canvas standalone components and blocks, optionally filtered by category. Design-system components (Andromeda and others) are not here: use list_systems. Use after list_categories to drill into one, or pass no filter to browse everything. Each result includes slug, name, description, categories, screenshot URL, install command, and homepage URL, enough to evaluate without a separate get_component call. Pagination via limit and offset.

Inputs: limit, offset, category.

list_systemsList AI Canvas design systemsA
Read-onlyIdempotent
Inspect

Return every design system available on AI Canvas. A design system is a coordinated set of components plus shared tokens and utilities, installable in one CLI command. Use when the user asks about "themes", "design systems", or wants more than a single component.

Inputs: no named parameters.

list_templatesList AI Canvas design-system templatesA
Read-onlyIdempotent
Inspect

Return every ready-made template on AI Canvas. A template is a complete example screen built from one design system (a dashboard, a console, a control room), installable in one CLI command. Use to discover what full screens exist before fetching one with get_template, or when the user asks for "a dashboard" / "an admin screen" rather than a single component.

Inputs: system.

search_componentsSearch AI Canvas componentsA
Read-onlyIdempotent
Inspect

Fuzzy keyword search across standalone components, design-system components, AND ready-made templates (whole example screens). Matches on slug, name, description, categories, and tags; returns best matches ranked by relevance. Use when the user describes what they want in their own words, e.g. "an animated card stack", "background with waves", "a mission control dashboard". Results include screenshot URLs and install commands for immediate evaluation. A result tagged "kind": "template" is a full screen. Fetch it with get_template, not get_component.

Inputs: query (required), limit.

search_iconsSearch AI Canvas iconsA
Read-onlyIdempotent
Inspect

Keyword search across the AI Canvas icon set (React icons, four styles each: Line is free with an account; Duotone, Solid and Solid duotone are Premium). Matches on name, tags and category, best match first. Use when the user needs an icon ("a trash icon", "something for settings") or when building UI that needs icons. Then call get_icon_install with the chosen name and style.

Inputs: query (required), limit, category.

validate_usageValidate AI Canvas component usage in a fileA
Read-onlyIdempotent
Inspect

Check a file's JSX against the documented props of the AI Canvas components it imports. Reports props that do not exist on the component, required props that are missing, and string values outside a prop's allowed options, each with a line number. Run it on every file that uses AI Canvas components before you finish. Components are matched by import path (components/aicanvas/ for standalones, components/aicanvas//... for design-system components); pass tags to map a tag to a slug by hand.

Inputs: code (required), path, tags.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 14 tools

    Disambiguation4/5

    The browse/fetch split is well handled: list_categories, list_components, list_systems, list_templates, and search_components each have an explicit scope, and descriptions actively redirect (e.g. list_components says design-system components are not here, use list_systems). The main friction is between get_component, get_component_props, get_system, and get_template, which all return overlapping 'files' or 'props' content and require the agent to reason about granularity rather than purpose. get_install_command vs get_icon_install is also a mild near-pair, distinguished only by component vs icon.

    Naming Consistency4/5

    Nearly every tool is verb_noun snake_case (list_components, search_icons, get_template, validate_usage), so the set reads predictably. Minor deviations: get_install_command and get_icon_install put the object in different positions and use different nouns for the same 'install command' concept, and validate_usage / get_audit_checklist lack a clear noun object. Still, no camelCase/snake_case mixing.

    Tool Count5/5

    14 tools sits squarely in the healthy range for a component-library server that must support discovery, inspection, installation, and validation. Each tool maps to a distinct step in the workflow and none appear to be redundant padding.

    Completeness4/5

    The lifecycle is well covered: orient (list_categories/list_systems/list_templates), discover (search_components), inspect (get_component/get_component_props/get_system/get_template), install (get_install_command/get_icon_install), and verify (validate_usage/get_audit_checklist). The notable gap is icons: search_icons and get_icon_install exist, but there is no way to fetch or preview an icon's source without installing, unlike the component path. Category-level browsing for systems and templates is also absent (only full lists).

    Maintenance

    ActivityActive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      D
      quality
      D
      maintenance
      A Model Context Protocol server that provides access to Magic UI components, allowing AI assistants and other MCP clients to discover and use UI components from the Magic UI design system.
      2
      14
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      A Model Context Protocol server that enhances AI agents by providing deep semantic understanding of codebases, enabling more intelligent interactions through advanced code search and contextual awareness.
      91
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      A Model Context Protocol server that provides AI assistants with access to ReactBits.dev components - a collection of 135+ animated React components for creative developers.
      5
      536 npm
      72
      MIT