Insider Design System MCP
Automated MCP server for the Insider Design System. Enables AI assistants to discover, understand, and generate code for over 60 Design System components with automated extraction from source code.
README
๐จ Insider Design System MCP
Automated Model Context Protocol server for the Insider Design System. Enables AI assistants like Claude to discover, understand, and generate code for 60+ Design System components with automated extraction from source code.
Version: 2.0 (Automated Extraction) Status: โ Production Ready Last Updated: 2025-11-21
โจ Features
๐ค Automated Extraction
- Zero Manual Work: Automatically extracts component metadata from Vue source files
- Always Up-to-Date: Re-run extraction when Design System changes (~5 minutes)
- Rich Metadata: Props, emits, enums, validators, slots - all extracted automatically
๐ Comprehensive Data
- 62 Components: Full coverage of Insider Design System
- 1,087 Props: With types, defaults, and validators
- 30 Enums: STYLES, TYPES, SIZES automatically detected
- Real Usage Analysis: Common mistakes detected from analytics-fe codebase
- Manual Enrichments: Critical components have detailed examples and notes
๐ง MCP Tools
list-components- List all componentsget-component- ๐ Markdown format - Get component info in human-readable format with 77% token savingssearch-components- Search by name/descriptiongenerate-code- Generate Vue component codemap-figma-component- Map Figma to DS components
โก Markdown Format (NEW!)
The get-component tool now returns component documentation in Markdown format for massive token savings:
Benefits:
- ๐ฐ 77% token savings compared to JSON format (690KB โ 161KB)
- ๐ Human-readable format with clear structure
- ๐ฏ ~135,335 tokens saved across all components
- โก Faster responses with smaller payloads
Top Performers:
- InButtonV2: 88% savings (55KB โ 6.6KB)
- InDropdownMenu: 87% savings (36KB โ 4.8KB)
- InTooltipV2: 86% savings (31KB โ 4.2KB)
Format includes:
- Props with types, defaults, descriptions
- Events with payloads
- Examples with code snippets
- Common mistakes and best practices
- Related components
๐ MCP Resources
ds://components- All components listds://registry- Registry metadatads://component/{name}- Individual componentds://categories- Component categories
๐ Documentation
โ See docs/ for complete documentation index
-
Architecture - System design and data flow
- How It Works - Complete architecture overview
- Smart Filter Layer - Token optimization system
- Figma Integration - Figma to Vue workflow
-
Guides - How-to guides and workflows
- Developer Workflow - Day-to-day development
- Enrichment Strategy - Creating enrichments
- Enrichment Template - Enrichment templates
-
Reference - API and tool references
- Agent Usage - Specialized agents guide
๐ Quick Start
Prerequisites
- Node.js >= 20.0.0
- npm >= 10.0.0
Installation
# Clone repository
git clone <repo-url>
cd design-system-mcp
# Install dependencies
npm install
# Extract component metadata (first time)
npm run extract:all
# Build
npm run build
# Test
npm run test:production
Need help with extraction scripts? See WORKFLOW.md for detailed usage guide.
โ๏ธ Configuration
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"design-system": {
"command": "node",
"args": ["/Users/YOUR_USERNAME/path/to/design-system-mcp/dist/index.js"]
}
}
}
Environment Variables
# Design System source path (for extraction)
export DS_PATH="/Users/YOUR_USERNAME/path/to/insider-design-system"
# Analytics FE path (for usage analysis)
export ANALYTICS_FE_PATH="/Users/YOUR_USERNAME/path/to/analytics-fe"
๐ Data Extraction Pipeline
Architecture
Design System Source Code (Vue files)
โ AUTOMATED EXTRACTION
data/combined.json (209 KB)
โ BUILD & COPY
dist/data/combined.json
โ RUNTIME LOADING
MCP Server (in-memory, enum-resolved)
โ FAST QUERIES
Claude Code
Extraction Commands
# Extract all data (run when Design System changes)
npm run extract:all
# Or run individually:
npm run extract:components # Parse Vue components โ data/components.json
npm run extract:storybook # Extract examples โ data/storybook.json
npm run extract:usage # Analyze usage โ data/usage.json
npm run extract:argtypes # Sync possibleValues from storybook argTypes
npm run extract:merge # Merge all โ data/combined.json
# Rebuild MCP server
npm run build
What Gets Extracted?
1. Component Metadata (extract-components.ts)
- Props (type, default, required, validator)
- Emits (from
$emit()calls) - Enums (const STYLES = {...})
- Slots
- Version (V1/V2)
2. Storybook Examples (extract-storybook.ts)
- Code examples from stories
- Descriptions
- Categories
3. Real Usage Analysis (extract-usage.ts)
- Usage counts from analytics-fe
- Common mistakes (auto-detected)
- Most used props
- Real code patterns
4. Manual Enrichments (overlay)
- Detailed valueFormat for critical props
- Common mistakes documentation
- Helper functions
- Migration guides
๐ Usage Examples
๐ฏ Complete Walkthrough: From Raw Request to Implementation
Scenario: You need a button that opens a dropdown menu.
Step 1: Optimize Your Raw Request
# In Claude Code (works in ANY project with Design System MCP!)
User: "Use optimize-prompt to transform this request: I need a button that opens a dropdown menu"
Claude calls mcp__design-system__optimize-prompt and returns:
# Original Prompt
I need a button that opens a dropdown menu
# Analysis
- Language: English
- Intent: implement
- Components: InButtonV2, InDropdownMenu
- Complexity: medium
- Estimated Token Savings: ~24,000 tokens
# Optimized Prompt
I need to implement InButtonV2 and InDropdownMenu components with proper integration.
Please use the Design System MCP to:
1. Get documentation for both components
- Tool: mcp__design-system__get-component
- Components: InButtonV2, InDropdownMenu
- Note: Markdown format saves ~87% tokens on average
2. Generate implementation code
- Tool: mcp__design-system__generate-code
- Include integration between components
Requirements:
- Primary styling for main actions
- Dropdown opens below trigger element
- Proper state management between components
Show me:
โ
Which components you're using
โ
Token savings from Markdown format for each component
โ
The generated Vue code
โ
Integration example showing how components work together
Step 2: Use the Optimized Prompt
User: "Now use the optimized prompt above to implement the button + dropdown"
Claude executes:
-
Gets InButtonV2 docs (Markdown format)
- Original: 55KB JSON โ 6.6KB Markdown
- Token savings: 88% (~12,198 tokens)
-
Gets InDropdownMenu docs (Markdown format)
- Original: 36KB JSON โ 4.8KB Markdown
- Token savings: 87% (~8,214 tokens)
-
Generates integrated Vue code:
<template>
<div class="button-dropdown-container">
<InButtonV2
id="actions-button"
:type="TYPES.PRIMARY"
:styling="STYLES.SOLID"
label-text="Actions"
@click="toggleDropdown"
/>
<InDropdownMenu
v-if="isOpen"
:items="menuItems"
@item-click="handleItemClick"
@close="isOpen = false"
/>
</div>
</template>
<script setup lang="ts">
import { ref } from 'vue';
import { InButtonV2, InDropdownMenu } from '@useinsider/design-system-vue';
import '@useinsider/design-system-vue/dist/design-system-vue.css';
const isOpen = ref(false);
const menuItems = [
{ id: 'export', label: 'Export Data', icon: 'download' },
{ id: 'share', label: 'Share Report', icon: 'share' },
{ id: 'settings', label: 'Settings', icon: 'settings' },
];
const toggleDropdown = () => {
isOpen.value = !isOpen.value;
};
const handleItemClick = (item: any) => {
console.log('Clicked:', item.id);
isOpen.value = false;
};
</script>
Summary
Total token savings: ~20,412 tokens (87.5% reduction) Time saved: Claude uses the right tools in the right order Quality: Proper integration, best practices included
๐ก Pro Tips
Always optimize raw requests first:
โ Don't do this:
User: "I need a button"
Claude searches through code files, wastes time.
โ Do this:
User: "Use optimize-prompt: I need a button"
Claude gets optimized prompt โ Uses MCP tools โ Fast & accurate implementation.
Works everywhere:
- โ analytics-fe project
- โ marketing-web project
- โ customer-portal project
- โ ANY project with Design System MCP configured
๐ฏ Quick Reference: optimize-prompt
// Single component
mcp__design-system__optimize-prompt("I need a button")
// Multiple components
mcp__design-system__optimize-prompt("button and dropdown menu")
// Migration task
mcp__design-system__optimize-prompt("migrate InDatePicker from V1 to V2")
// Debug issue
mcp__design-system__optimize-prompt("InSelect not working, showing errors")
// Learning
mcp__design-system__optimize-prompt("how to use InTooltipV2?")
Get Component Details
// Claude automatically calls:
mcp__design-system__get-component("InButtonV2")
// Returns in Markdown format (88% token savings):
# InButtonV2
**Version:** v2
## Props
### `styling`
**Type:** `String` | **Default:** `"STYLES.SOLID"`
**Allowed values:** `solid`, `ghost`, `text`
### `type`
**Type:** `String` | **Default:** `"TYPES.PRIMARY"`
...
## Examples
### Basic Primary Button
```vue
<InButtonV2
id="primary-btn"
styling="solid"
type="primary"
label-text="Click Me"
/>
Token savings: 55KB โ 6.6KB (88% reduction)
### Search Components
```typescript
mcp__design-system__search-components("button")
// Returns: InButton, InButtonV2, InCreateButton...
Generate Code
mcp__design-system__generate-code({
component: "InButtonV2",
props: { styling: "solid", type: "primary" }
})
// Returns:
// <InButtonV2
// id="button-1"
// styling="solid"
// type="primary"
// label-text="Button"
// />
๐งช Testing
# Test combined dataset
npm run test:data
# Test production build
npm run test:production
# Run unit tests
npm test
# Coverage
npm run test:coverage
๐ Project Structure
design-system-mcp/
โโโ ๐ README.md # This file
โโโ ๐ COMPLETION_REPORT.md # Full project report
โโโ ๐ HOW_IT_WORKS.md # Architecture deep dive
โโโ ๐ CLEANUP_SUMMARY.md # Cleanup history
โโโ ๐ฆ package.json # Dependencies & scripts
โโโ ๐ง tsup.config.ts # Build configuration
โ
โโโ ๐ src/ # Source code
โ โโโ index.ts # Entry point
โ โโโ server.ts # MCP server
โ โโโ tools/index.ts # MCP tools
โ โโโ resources/index.ts # MCP resources
โ โโโ types/index.ts # TypeScript types
โ โโโ registry/
โ โโโ combined-loader.ts # โญ Dataset loader (NEW)
โ โโโ enrichments/ # Manual enrichments
โ โ โโโ InButtonV2.json
โ โ โโโ InDatePickerV2.json
โ โ โโโ InSelect.json
โ โโโ migrations/ # V1โV2 guides
โ โโโ InDatePicker-to-V2.json
โ
โโโ ๐ scripts/ # Extraction scripts
โ โโโ extract-components.ts # Vue component parser
โ โโโ extract-storybook.ts # Example extractor
โ โโโ extract-usage.ts # Usage analyzer
โ โโโ merge-datasets.ts # Dataset combiner
โ
โโโ ๐ data/ # Extracted data
โ โโโ components.json # 148 KB - Parsed components
โ โโโ storybook.json # 1.6 KB - Examples
โ โโโ usage.json # Real usage data
โ โโโ combined.json # 209 KB - โญ FINAL DATASET
โ
โโโ ๐ dist/ # Build output
โโโ index.js # Bundled MCP server
โโโ data/combined.json # Runtime dataset
๐ Update Workflow
When Design System Changes
# 1. Pull latest Design System
cd /path/to/insider-design-system
git pull
# 2. Re-extract metadata
cd /path/to/design-system-mcp
npm run extract:all # ~5 minutes
# 3. Rebuild MCP server
npm run build
# 4. Test
npm run test:production
# 5. Commit (optional)
git add data/combined.json
git commit -m "chore: update component metadata"
git push
# Claude Desktop will auto-reload! โ
Before: 2-3 hours manual work Now: 5 minutes automated! ๐
For more scenarios and detailed workflow guide, see WORKFLOW.md.
๐ก Key Innovations
1. Automated Extraction
No more manual JSON editing. Parser reads Vue files directly.
2. Enum Resolution
// Source: const STYLES = { SOLID: 'solid', GHOST: 'ghost' }
// Extracted: enums: [{ name: "STYLES", values: {...} }]
// Runtime: validValues: ["solid", "ghost", "text"] โ
3. Real Usage Intelligence
Scans analytics-fe for common mistakes:
{
"mistake": "Using number for iconSize",
"occurrences": 12,
"fix": "Use string: icon-size=\"24\"",
"severity": "critical"
}
4. Layered Enrichment
Auto-extracted (100% coverage)
+
Manual enrichments (critical details)
=
Best of both worlds! โ
๐ Data Quality
Components: 62 (100% coverage)
Props: 1,087 (with types, defaults, validators)
Enums: 30 (automatically detected)
Emits: 170
Manual Enrichments: 3 (InButtonV2, InDatePickerV2, InSelect)
Migration Guides: 1
File Size: 209 KB (combined.json)
๐ฏ Benefits
For Developers
- โ Accurate component information
- โ Enum values always correct
- โ Common mistakes documented
- โ Real usage examples
- โ Fast code generation
For Design System Team
- โ Zero manual maintenance
- โ Always synchronized with source
- โ Easy updates (5 minutes)
- โ Automatic mistake detection
Expected Impact
- Code Generation Accuracy: 30% โ 85%
- First-Try Correctness: 20% โ 80%
- Onboarding Time: -70%
- Design System Questions: -50%
๐ ๏ธ Development
Scripts
# Build
npm run build # Build for production
npm run dev # Watch mode
# Extraction
npm run extract:components # Extract component metadata
npm run extract:storybook # Extract examples
npm run extract:usage # Analyze real usage
npm run extract:merge # Merge all datasets
npm run extract:all # Run all extractions
# Testing
npm test # Run unit tests
npm run test:coverage # Coverage report
npm run test:data # Test dataset validity
npm run test:production # Test production build
# Code Quality
npm run lint # Run ESLint
npm run lint:fix # Fix linting issues
npm run typecheck # TypeScript check
Adding New Enrichments
Option 1: Use enrichment-maker agent (Recommended)
# Let AI generate enrichment for you
# In Claude Code: "Use enrichment-maker agent to create enrichment for InTooltipV2"
# Agent analyzes component and creates:
# - valueFormat for complex props
# - commonMistakes documentation
# - Real-world examples
# - Helper functions
# Then merge and build:
npm run extract:merge
npm run build
Option 2: Manual creation
# 1. Create enrichment file
touch src/registry/enrichments/InTooltipV2.json
# 2. Add detailed metadata
# (See existing enrichments: InButtonV2, InDatePickerV2, InSelect)
# 3. Rebuild
npm run extract:merge
npm run build
๐ Documentation
- README.md (this file) - Quick start & overview
- WORKFLOW.md - โญ When and how to run extraction scripts
- AGENT_USAGE.md - ๐ค How to use agents and slash commands
- HOW_IT_WORKS.md - Architecture deep dive
- COMPLETION_REPORT.md - Full project report
- CLEANUP_SUMMARY.md - Cleanup history
- CLAUDE.md - Instructions for Claude Code
๐ค Contributing
- Fork the repository
- Create feature branch:
git checkout -b feature/amazing - Make changes
- Run tests:
npm test - Commit:
git commit -m "feat: add amazing feature" - Push:
git push origin feature/amazing - Submit Pull Request
๐ License
UNLICENSED - Internal use only (Insider)
๐ฌ Support
For questions and issues:
- Create GitHub issue
- Contact Design System team
- Slack: #design-system
๐ Success Stories
"Component metadata is now always accurate. Claude generates correct code on first try!" โ Developer using MCP
"We updated 15 components in Design System. Re-extraction took 5 minutes!" โ Design System Team
Built with โค๏ธ by the Insider Design System Team
Powered by: Claude Code (Sonnet 4.5)
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.