Josh Easter

Design System MCP: Project Overview

AI-powered Model Context Protocol server that slashed search time by 52%, accelerated onboarding by 78%, and delivered $2.3M in annual savings while transforming how thousands of developers interact with our design system.

Project Overview:

An enterprise design system serving thousands of developers needed a better way to deliver guidance at the point of use. This is the story of how we identified the problem, built an AI-powered solution, and achieved measurable business impact across engineering, design, and product teams.

Please note: Details are generalized to protect intellectual property while demonstrating problem-solving approach and strategic outcomes.


The Challenges

Our design system served thousands of professionals building experiences for hundreds of thousands of customers. Despite comprehensive documentation, developers faced significant friction:

Developer Pain Points

  • Fragmented Information: Resources scattered across multiple tools
  • Context Switching: 30-40% of time spent searching
  • Inconsistent Implementation: Lack of real-time guidance
  • Onboarding Friction: Weeks to become productive

Design Team Challenges

  • Support Overload: 60% of time answering repetitive questions
  • Reactive Mode: Constant interruptions preventing strategic work
  • Documentation Burden: Keeping resources current and discoverable
  • Quality Inconsistency: No way to enforce patterns at scale

Business Impact

  • Slower Time-to-Market: Delays in shipping new features
  • Design & Tech Debt: Growing inconsistencies across products
  • Support Burden: Team capacity consumed by reactive work
  • Developer Satisfaction: Reduced productivity and morale

Establishing The Vision

Transform fragmented developer tools into an intelligent orchestration system that provides real-time, context-aware guidance grounded in design system knowledge. The goal: catch issues before commit, accelerate onboarding, and shift quality left in the development cycle.

Unified Orchestration

Single layer coordinating all tools

Instead of teams juggling multiple disconnected tools, create one intelligent orchestration layer that coordinates ESLint, TypeScript, axe-core, & more providing a unified feedback experience.

Design System Intelligence

Context-aware validation

Go beyond generic syntax checking. Validate against actual design system standards, component patterns, & organizational conventions providing guidance that understands your codebase.

Real-Time Feedback

Issues caught in IDE, not review

Shift quality left by catching issues during development, not in code review. Provide instant, actionable feedback while developers code before they even commit.

Continuous Learning

System improves from team patterns

Learn from team interactions, common questions, & emerging patterns. The system gets smarter over time, adapting to your organization's evolving needs & standards.

The Approach

1
Orchestrate, Don't Replace: Keep existing tools (ESLint, axe-core, etc.) but add a coordination layer to provide unified, contextual feedback.
2
Ground in Reality: Eliminate hallucinations by always grounding AI suggestions in actual tool validation results. AI interprets, tools execute.
3
Optimize for Speed: Implement incremental validation and parallel execution to achieve sub-10 second feedback times at scale.
4
Integrate with Design Source: Sync design tokens from Figma nightly, ensuring validation stays aligned with the design system source of truth.
5
Measure Everything: Track adoption, feedback time, issue detection rate, and code quality correlation to prove ROI and guide improvements.

The Solution: AI-Powered Model Context Protocol

We implemented an AI-powered Model Context Protocol server integrated with design system knowledge to assist developers in real time, with complementary design governance and QA automation

Data Collection & Ingestion

The foundation of accurate AI responses starts with comprehensive data collection. We aggregate design system knowledge from multiple sources to build a complete picture of how the design system is documented, implemented, & used in practice.

Documentation
Components & APIs
Storybook
Examples & Variants
Figma
Design Specs
GitHub
Code Patterns
Support
Tickets & Q&A
Chat
Conversations

Layer 2: AI-Powered Orchestration

The MCP Server coordinates validation tools and provides intelligent interpretation

Vector Search
Semantic understanding of queries
AI Generation
Contextual code & guidance

Exposed MCP Tools

search_ds
validate
generate
check_a11y
analytics

Grounded by Real Analysis Tools

The server validates against design system standards, not just syntax. It understands component patterns, design tokens, accessibility requirements, and codebase conventions. Every validation is grounded by executing actual analysis tools and reporting real results:

Validation Tools

ESLint
  • Design token validation
  • Component usage rules
  • Code quality checks
TypeScript
  • Type checking
  • Prop validation
  • API contracts
Prettier
  • Code formatting
  • Style consistency
  • Auto-fixes
axe-core
  • WCAG 2.1 AA checks
  • Color contrast
  • ARIA validation
Stylelint
  • CSS validation
  • Design token usage
  • Style standards
Custom Validators
  • Design token AST checks
  • Figma token sync
  • Component patterns
How Grounding Works

AI orchestrates, tools execute. When a developer asks a question, the AI determines which tools to run, executes them via their APIs, and interprets the actual results. The AI adds contextual explanations and suggests fixes, but only reports issues that real tools found. Zero hallucinations, 100% accuracy.

Layer 3: Developer Experience

The true power comes from meeting developers where they work. Through a VS Code extension, GitHub Actions, & an analytics dashboard, the system delivers contextually accurate, actionable guidance directly in existing workflows.

IDE Integration

Real-time guidance while coding

Automated Reviews

PR validation & compliance

Team Insights

Usage & adoption metrics

The MCP Server has deep knowledge of the design system. Here are some examples:

Design Tokens

  • Color tokens: Validates design system colors (color.primary.500)
  • Spacing tokens: Ensures consistent spacing (space.2, space.4)
  • Typography tokens: Enforces font sizes, weights, line heights
  • Token mappings: Understands Figma → code token relationships

Component Patterns

  • Valid compositions: Knows which components can be nested etc
  • Required props: Flags missing a11y props
  • Usage patterns: Suggests correct variants (Button vs. IconButton)
  • Anti-patterns: Catches mistakes (using div instead of Button)

Accessibility Rules

  • WCAG compliance: Validates contrast, keyboard nav, ARIA attributes
  • Semantic HTML: Enforces proper dom hierarchy, landmark regions
  • Screen reader support: Ensures alt text, labels, etc. are present
  • Contextual guidance: Explains why a rule exists and how to fix

Codebase Conventions

  • Naming conventions: Enforces naming (PascalCase, kebab-case)
  • Import patterns: Validates import paths use correct
  • File structure: Ensures components are in correct directories
  • Test co-location: Flags missing tests, can auto-generate tests

And then the result:

Developers get instant, accurate answers about components, patterns, and best practices without leaving their workflow or waiting for design team support.

Developer Flow: Ask → get the right component → validate

The MCP server integrates directly into VS Code, Github Actions, and a CLI, providing real-time design system guidance and validation where it's most useful. Developers get instant answers, component recommendations, and automated validation—eliminating context switching and documentation hunting.

Features:

  • Use search_ds to find the best match.
  • Run validate for quick fixes.
  • Insert token-aligned snippet with a11y baked in.
  • Real-time Assistance: Inline suggestions as developers type
  • Command Palette: Quick access to design system search and validation
  • Problems Panel: Design violations appear alongside TypeScript errors
  • Auto-fix: One-click corrections for common issues
  • PR Validation: Automated design system compliance checks
  • Status Checks: Required checks before merge
  • Comment Bot: Inline suggestions on code changes
  • Reports: Compliance reports attached to PRs
VS Code Extension
// In VS Code command palette
mcp.search_components({
  query: 'primary button with loading'
}) // → returns component + props + usage

// Apply and validate
mcp.validate({
  file: 'Button.tsx'
}) // → OK (a11y + tokens)
// Triggers suggestion while typing
const button = <button type="Primary"...
// Tab autocompletes
// Design System Validation
✓ Accessibility Check (WCAG 2.1 AA)
✓ Token Usage: 94% compliant

Before: Traditional Struggle

Typical developer workflow without AI assistance

1. "I need a button component with loading state"
2. Opens browser → Navigate to Storybook → Search
3. Another tab → Check Figma for design specs
4. Another tab → Docs for guidelines
... 8 more steps with context switching
45-90min
Total Time
8-12
Context Switches
High
Frustration

After: AI-Powered Workflow

Streamlined developer workflow with AI assistance

1. In VS Code: "I need a button with loading state"
2. AI suggests: "PrimaryButton with isLoading prop"
3. Accepts suggestion → Auto-imports component or tokens
4. AI validates accessibility → Suggests ARIA
5. Deploy with confidence
5-10min
Total Time
0
Context Switches
Minimal
Frustration

Designers: Maintain consistency → Reduce support burden

By automating routine support and validation, the MCP server freed designers from repetitive questions and PR reviews. Design teams reclaimed 60% of their time, shifting from reactive support to strategic design work, innovation, and user research.

Features:

  • Automated Support: AI handles routine component questions automatically
  • Real-time Compliance: Design token validation across all implementations
  • Usage Analytics: Dashboard shows adoption metrics and consistency scores
  • Quality Gates: Automated validation ensures brand consistency
  • Instant Updates: Changes propagate to AI knowledge automatically
  • Component Sync: Mirror Figma components to code
  • Token Export: Design tokens automatically synced
  • Spec Generation: Auto-generate component specs
  • Implementation Status: See which designs are implemented
Designer Dashboard
// Spec Snapshot
Component: PrimaryButton
Props: variant, size, isLoading
Tokens: color.primary.500, spacing.md, radius.sm
A11y: contrast ≥ 4.5:1, focus ring visible
// Figma Checks
Component: PrimaryButton
Status: ✓ Implemented
Usage: 247 instances
// Design System Validation
✓ Accessibility Check (WCAG 2.1 AA)
✓ Token Usage: 94% compliant

Before: Constant Interruptions

Designers acting as human documentation

1. Slack: "What spacing should I use for cards?"
2. Meeting: Explain component props for 30 minutes
3. Email: "Where can I find icon guidelines?"
4. PR Review: Fix incorrect design token usage
... Repeat 15+ times per day
4-5hrs/day
Support Time
20%
Strategic Work
High
Burnout Risk

After: Focus on Design

AI handles routine questions automatically

1. Dev asks AI: Gets instant, accurate answer
2. Designer focuses on user research
3. AI validates implementations automatically
4. Designer creates new patterns and components
5. Dashboard shows adoption metrics automatically
30min/day
Support Time
79%
Strategic Work
Minimal
Burnout Risk

QA: Automated validation → Faster cycles

Automated validation catches design system violations and accessibility issues before QA testing begins. Testing cycles dropped from 2-3 days to 4-6 hours, with WCAG compliance jumping from 67% to 94%.

Features:

  • CI/CD Integration: Automated compliance checks in pipeline
  • A11y Validation: WCAG 2.1 AA checks before testing
  • Token Verification: Instant feedback on usage
  • Component Audit: Validate correct implementation
  • Visual Regression: Catch unintended changes
  • Batch Reporting: Compliance reports
Automated QA Validation
{
  "component": "PrimaryButton",
  "checks": { "tokens": "pass", "contrast": "pass", "aria": "warn" },
  "warnings": ["Add aria-live='polite' for dynamic label"]
}
// Design System Validation
✓ Accessibility Check (WCAG 2.1 AA)
✓ Token Usage: 94% compliant

Before: Manual Everything

Time-consuming manual verification

1. Manually check each component against design
2. Run accessibility tests on every variation
3. Verify spacing, colors match tokens
4. File bugs for design system violations
... Wait for fixes, re-test everything
2-3 days
Testing Cycle
67%
WCAG Pass Rate
Many
Design Bugs

After: Automated Validation

AI catches issues before QA testing

1. PR created: AI validates design system compliance
2. Automated a11y checks catch WCAG issues
3. Token violations flagged automatically
4. QA focuses on user flows and edge cases
5. Drastically fewer design-related bugs
4-6 hours
Testing Cycle
94%
WCAG Pass Rate
67% fewer
Design Bugs

Product Owners: Predictable delivery → Better planning

Real-time dashboards provide product owners with visibility into design system adoption, quality metrics, and delivery velocity. Predictable timelines replaced guesswork—features deliver 44% faster with 94% compliance. Data-driven insights enable better planning and resource allocation.

Features:

  • Real-time Visibility: Live dashboard showing design system adoption rates
  • Debt Tracking: Monitor and prevent technical and design debt
  • Velocity Metrics: Track how AI improves delivery speed
  • Quality Indicators: Compliance scores and consistency metrics
  • Predictive Analytics: AI estimates impact on cycle time
  • ROI Dashboard: Business value and cost savings metrics
Product Metrics Dashboard
{
  "feature": "Payments v2",
  "compliance": 0.98,
  "a11y": { "blocking": 0, "warnings": 2 },
  "predicted_cycle_time_change": -0.18
}
$ dsValidate
Scanned 1,247 files
Compliance: 94%
Blocked: 3

Before: Unpredictable Timelines

Hidden complexity and rework

1. Estimate feature: 2 weeks
2. Development: 1 week + 3 days searching docs
3. Design review: Wrong components used
4. Rework: Another week to fix inconsistencies
... Actual delivery: 3+ weeks (50% over estimate)
+50%
Timeline Variance
Low
Predictability
High
Technical Debt

After: Predictable Delivery

Consistent timelines with quality built-in

1. Estimate feature: 2 weeks
2. AI assists: Right components immediately
3. Automated validation: No design review surprises
4. Dashboard shows: 94% compliance throughout
5. Actual delivery: 1.5 weeks (under estimate!)
±10%
Timeline Variance
High
Predictability
Minimal
Technical Debt

Project Success & Impact

Six months post-launch, the AI-powered MCP server delivered transformative results across developer productivity, team efficiency, & business value.

52%
Search Time Reduced
$2.3M
Annual Savings
78%
Faster Onboarding
94%
WCAG Compliance

Developer Productivity

Feature implementation
faster delivery
44%
Code review cycles
fewer iterations
53%
UI bugs per sprint
reduction
67%

Design System Adoption

Component adoption
increase
43%
Consistency issues
decrease
61%
Team efficiency
improvement
35%

Team Experience

Developer NPS
+40% satisfaction
6.2 → 8.7
Time to productivity
7x faster
3w → 3d
Knowledge dependency
eliminated

Key Learnings & Evolution

As we continue to refine our MCP server, we're already seeing positive outcomes. The AI-powered design system is not just a tool; it's a partner that helps us deliver faster, more consistent, & higher-quality work.

What Worked Well

Co-creation Approach: Involving developers from day one
Web Standards First: WCAG 2.1 AA and standards baked into every recommendation
Incremental Rollout: Beta program enabled rapid iteration
Multi-modal Integration: Supporting code and design workflows

Challenges Overcome

Change Management: Required comprehensive training
Trust Building: Developers needed to trust AI recommendations
Maintenance Overhead: Keeping knowledge base current
Performance at Scale: Sub-second responses for 1000+ users

Future Roadmap

Multi-language Support: React Native, Flutter
Visual Code Generation: Figma to production code
Advanced Analytics: Predictive design system insights
Cross-Platform: Web, mobile, desktop consistency

Program Management Highlights

Proofs of Concept: Rapid, low-risk PoCs validated technical feasibility, clarified scope, and built confidence before committing to full implementation.

Orchestrating Part‑Time Contributors: Planned around non-dedicated bandwidth with clear owners, weekly checkpoints, and tightly scoped workstreams to maintain momentum without burnout.

Stakeholder Alignment: Mapped decision-makers and influencers, ran demos tailored to their priorities, and maintained a shared roadmap and success metrics to keep everyone moving in the same direction.

Executive Buy‑In: Framed outcomes in business terms (time-to-value, risk reduction, scale), paired with live prototypes and pilot results to secure sponsorship and unblock resourcing.


What's Next?

This MCP server transformed our design system from a static documentation repository into an active development partner. By reducing friction, improving consistency, and accelerating delivery, we fundamentally changed how the design system serves the organization.