Lets an AI client look up a design system's components, slots and presets and validate screens against it via MCP.
Also known as 0design/ai-native-design-framework, @ai-native-design-framework/kit
- Stars
- 29
- Tools
- —
- Downloads, 30 days
- 177
- Latest version
- —
- Last commit
- Nov 4, 2026
- License
- MIT
What is AI-Native Design Framework MCP?
AI-Native Design Framework lets an AI client look up a design system's components, slots and presets and validate screens against it via MCP.
Summary generated from the project's README
About AI-Native Design Framework MCP server
Preventing AI agents from hardcoding UI and ignoring design systems is the primary goal of this framework. It allows developers to connect their specific design system components, slots, modifiers, and presets to an AI client via a standard MCP connection. By using this server, an agent can look up exactly what a design system offers instead of writing custom CSS or HTML that deviates from established brand guidelines.
To set this up, you first use the kit to check and bundle your design system files, such as tokens, taxonomy, and components, into a single JSON bundle. You then connect this bundle to Claude Code or any other agent using a stdio command. The workflow requires Node.js 20.10 or newer. Once connected, the agent generates screens as configurations that must pass the validate-screen tool to ensure they strictly adhere to the provided design system rules.
What you can do with it
- 1Enforce design system complianceInstruct an AI agent to build a new user interface screen using only the components and presets defined in your specific design system bundle.
- 2Validate UI configurationsAsk the AI client to run a validation check on a screen configuration to ensure no hardcoded values or unsupported properties are being used.
- 3Identify missing design tokensHave the agent report when a requested UI element requires a property or color that does not currently exist in the design system.
Written from the project's README. Every use case is checked against a quote from it.
How to install AI-Native Design Framework MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall AI-Native Design Framework MCP in Claude Code
$ claude mcp add ai-native-design-framework -- npx -y @ai-native-design-framework/kitInstall AI-Native Design Framework MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"ai-native-design-framework": {
"command": "npx",
"args": [
"-y",
"@ai-native-design-framework/kit"
]
}
}
}Install AI-Native Design Framework MCP in VS Code
.vscode/mcp.json
{
"servers": {
"ai-native-design-framework": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@ai-native-design-framework/kit"
]
}
}
}Install AI-Native Design Framework MCP in VS Code CLI
$ code --add-mcp '{"name":"ai-native-design-framework","type":"stdio","command":"npx","args":["-y","@ai-native-design-framework/kit"]}'Install AI-Native Design Framework MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"ai-native-design-framework": {
"command": "npx",
"args": [
"-y",
"@ai-native-design-framework/kit"
]
}
}
}Install AI-Native Design Framework MCP in Codex
$ codex mcp add ai-native-design-framework -- npx -y @ai-native-design-framework/kitInstall AI-Native Design Framework MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.ai-native-design-framework] command = "npx" args = ["-y", "@ai-native-design-framework/kit"]
Install AI-Native Design Framework MCP in Gemini CLI
$ gemini mcp add -s user ai-native-design-framework npx -- -y @ai-native-design-framework/kitInstall AI-Native Design Framework MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"ai-native-design-framework": {
"command": "npx",
"args": [
"-y",
"@ai-native-design-framework/kit"
]
}
}
}Automated checks
Ran Nov 13, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachable0design/ai-native-design-framework is public and reachablegithub.com/0design/ai-native-design-framework ↗Pass
- Package names this repository as its source@ai-native-design-framework/kit names 0design/ai-native-design-framework as sourcenpmjs.com/package/@ai-native-design-framework/kit ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for @ai-native-design-framework/kitnpmjs.com/package/@ai-native-design-framework/kit/v/0.2.1#pr ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-04github.com/0design/ai-native-design-framework ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from @ai-native-design-framework/kitPass
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub"You’re absolutely right, I ignored your design system and hardcoded UI. Again."— Your AI-Agent
Get started
Add AINDF to my design system and this project. AINDF is at https://github.com/0design/ai-native-design-framework; version 0.2 runs from a clone. Clone it to
<dir>(an absolute path). In my design system's folder, write my design system down likeexamples/demo-ds: the seven 0.2 sources (tokens,taxonomy,slots,applicability,presets,components,bindings) plusaindf.config.json, which lists them (schemas: 0.1 inschemas/, 0.2 inpackages/aindf-kit/schema/0.2). From that folder runnode <dir>/packages/aindf-kit/src/cli.mjs check aindf.config.jsonuntil it passes, thennode <dir>/packages/aindf-kit/src/cli.mjs bundle aindf.config.json --out aindf.bundle.json. Connect the bundle to your agent's MCP settings; in Claude Code runclaude mcp add aindf -- node <dir>/packages/aindf-kit/src/cli.mjs mcp <absolute path>/aindf.bundle.json(or start Claude Code with--plugin-dir <dir>/pluginand set its bundle path); any other agent runs the samenode … mcp …command as a stdio server in its MCP settings. From then on, write screens only as screen configs pinned to that bundle: look up components, slots, modifiers and presets, and callvalidate-screenuntil it passes. Never hardcode a value my design system does not have; when something is missing, tell me instead of writing your own CSS. Do not change the design system without my approval.
Install
The prompt above works from a clone; the kit is also on npm, so you can do the same without one. You need Node.js 20.10 or newer. From your design system's folder:
npx -y @ai-native-design-framework/kit check aindf.config.json
npx -y @ai-native-design-framework/kit bundle aindf.config.json --out aindf.bundle.json
claude mcp add aindf -- npx -y @ai-native-design-framework/kit mcp /absolute/path/to/aindf.bundle.json
The first command checks your design system, the second makes the bundle (one file per version), the third connects it to Claude Code. Any other agent runs the last command as a stdio MCP server.
To give an agent without the plugin the screen-authoring skill, install it with the skills CLI:
npx skills add 0design/ai-native-design-framework
It installs aindf-screen-author, which works through the MCP server connected above.
How it works
your prompt
│
▼
your agent ◀──── MCP ────▶ your design system
│ asks what it offers: components, slots, modifiers, presets
▼
screen config (JSON, pinned to one version of your design system)
│
▼
check ── refused: "Heading has no prop color; …" ──▶ your agent asks you
│ to add it to your
│ passes design system
▼
builder ──▶ a screen built only from your design system
Without AINDF your agent writes HTML and CSS itself and can ignore your design system. With AINDF the screen is a
config, and the check passes only what your design system offers. Fix how a component looks or works in your design
system once, and every screen built from it gets the fix on the next build, once the screens are moved to the new version (aindf repin).
Check a screen against your design system as a workflow step: aindf-validator on QFactory
What's in this repo
SPEC.md: the specification.packages/aindf-kit: the check, the builder and the MCP server.plugin: the plugin for Claude Code.examples/demo-ds: a demo design system with three examples you can run.schemasandpatterns: the JSON Schemas of version 0.1 and an example pattern. The 0.2 schemas are inpackages/aindf-kit/schema/0.2.
Contribute
You can improve the spec, the kit or the plugin, or suggest an idea.
- Fork the repository and make a branch.
- Make your change and run the checks:
node --test packages/aindf-kit/test/*.test.mjsandnode examples/demo-ds/examples.mjs. If you change a schema, updateSPEC.mdandpatterns/starter.jsonin the same pull request. - Open a pull request. Changes to the kit, the plugin and the examples are checked automatically.
Have an idea? Open an issue and describe what result you want.
Links
- Specification: SPEC.md
- Error codes: RULES.md
- Demo design system: examples/demo-ds
- For AI agents: build screens only through the design system's MCP. The rules are in the skill.
- License: MIT
AI-Native Design Framework MCP questions
How do I add AI-Native Design Framework MCP to Claude Code?
Run claude mcp add ai-native-design-framework -- npx -y @ai-native-design-framework/kit in your terminal. Add --scope user to make it available in every project.
How do I add AI-Native Design Framework MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
How do I connect the design system to Claude Code?
Bundle your design system into one file with the kit, then add it as a stdio MCP server using the claude mcp add command.
What does the MCP server let an agent do?
It lets the agent ask what the design system offers, such as components, slots, modifiers and presets, and validate screen configs against it.
How does an agent check a screen against the design system?
It writes the screen as a JSON screen config pinned to a bundle and calls validate-screen with the kit until the check passes.
What happens when the design system lacks something the agent needs?
The check refuses the screen and the agent should ask you to add it to the design system instead of hardcoding values.
What Node.js version is required?
You need Node.js 20.10 or newer to run the kit from npm or a clone.