Design: to-Code Workflow with Figma MCP

Design-to-Code Workflow with Figma MCP

Confidence: Tier 2 — Based on documented production case studies (Parallel HQ, builder.io), MCP server specifications, and community workflows.

Automated design system implementation using Figma MCP Server enables Product Designers to hand off production-ready specifications to Claude Code, which implements components maintaining 1:1 design-code parity.

Table of Contents

1. TL;DR
2. Documented Impact
3. Architecture Overview
4. 3-Tier Token Hierarchy
5. Prerequisites
6. Core Workflows
7. Code Connect Setup
8. Example Prompts
9. Team Adoption Patterns
10. Anti-Patterns
11. Implementation Roadmap
12. Resources

TL;DR

Designer (Figma Make) → Export (Figma Design) → Claude (Figma MCP) → Production Code

Key insight: Design system = source of truth
Claude consumes tokens/components directly from Figma
Implementation maintains design parity automatically

Documented Impact

Based on production case studies from January 2026:

Typical workflow timing:

• Single frame → Production component: 2-3 minutes
• Design system drift audit: 3 weeks → 3 minutes
• Token update propagation: Manual hours → Automated seconds

Sources: builder.io/blog/claude-code-figma-mcp-server, parallelhq.com/blog/automating-design-systems-with-ai, composio.dev/blog/how-to-use-figma-mcp-with-claude-code

Architecture Overview

Full Stack

[Figma Design File]
    ↓ (Variables & Styles)
[Tokens Studio Plugin] (optional but recommended)
    ↓ (JSON export)
[GitHub Repository]
    ↓ (CI/CD)
[Style Dictionary]
    ↓ (Transform)
[CSS Custom Properties / Tailwind Config]
    ↓ (Consumed by)
[Component Library]
    ↑ (Reads via)
[Claude Code + Figma MCP]

MCP Integration Point

Claude Code accesses Figma through the Figma MCP Server:

Claude Code
    ↓ (uses)
Figma MCP Server (mcp-server-figma)
    ↓ (authenticates via)
Figma Personal Access Token
    ↓ (reads)
Figma File (Dev Mode data)

What Claude can access:

• File structure and frames
• Color/text/effect styles
• Component properties
• Variables (tokens)
• Dev Mode annotations
• Code Connect snippets (if configured)

What Claude cannot access:

• Private files without token permissions
• Edit capabilities (read-only)
• Real-time collaboration data
• Version history (only current state)

3-Tier Token Hierarchy

Modern design systems use a hierarchical token structure. Claude Code understands this hierarchy when consuming Figma data.

Example Hierarchy

Base:
  --color-blue-600: #0066CC
  --spacing-2: 8px
  --radius-md: 4px

Composite:
  --button-padding: var(--spacing-2) var(--spacing-4)
  --button-border-radius: var(--radius-md)

Semantic:
  --interactive-primary: var(--color-blue-600)
  --interactive-primary-hover: var(--color-blue-700)

Claude Code behavior: When given a Figma component, Claude:

1. Extracts referenced variables (base tier)
2. Identifies composite patterns (spacing, sizing)
3. Applies semantic naming from your token conventions
4. Generates code matching this hierarchy

Prerequisites

For Designers

For Developers

MCP Configuration

Add to your Claude Code MCP settings (.claude/mcp.json or settings UI):

{
  "mcpServers": {
    "figma": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-figma"],
      "env": {
        "FIGMA_PERSONAL_ACCESS_TOKEN": "your-token-here"
      }
    }
  }
}

Security note: Use environment variables for production:

{
  "env": {
    "FIGMA_PERSONAL_ACCESS_TOKEN": "${FIGMA_TOKEN}"
  }
}

Then export in your shell: export FIGMA_TOKEN="figd_..."

Core Workflows

Workflow A: Single Frame → Production Component

Timing: 2-3 minutes per component

Steps:

1. Designer: Create component in Figma with proper variables/styles
2. Designer: Share Figma file URL with dev/Claude
3. Developer: Prompt Claude Code:

Read the "Button/Primary" component from Figma file:
https://www.figma.com/design/FILE_KEY

Implement as a React component with TypeScript.
Use Tailwind for styling, mapping Figma variables to our design tokens.
Ensure responsive behavior matches Figma's Auto Layout constraints.

4. Claude:

   • Fetches component via Figma MCP
   • Extracts styles, dimensions, spacing
   • Maps variables to code tokens
   • Generates component with props matching Figma variants

5. Verification: npm run dev → Visual comparison against Figma

Example output:

// components/Button/Primary.tsx
interface ButtonProps {
  size?: 'sm' | 'md' | 'lg';
  disabled?: boolean;
  children: React.ReactNode;
}

export function PrimaryButton({ size = 'md', disabled, children }: ButtonProps) {
  return (
    <button
      className={cn(
        'rounded-md font-medium transition-colors',
        'bg-interactive-primary text-white',
        'hover:bg-interactive-primary-hover',
        'disabled:opacity-50 disabled:cursor-not-allowed',
        {
          'px-3 py-1.5 text-sm': size === 'sm',
          'px-4 py-2 text-base': size === 'md',
          'px-6 py-3 text-lg': size === 'lg',
        }
      )}
      disabled={disabled}
    >
      {children}
    </button>
  );
}

Workflow B: Design System Drift Audit

Timing: 3 weeks manual review → 3 minutes automated

Problem: Over time, code diverges from design system (magic numbers, hardcoded colors, inconsistent spacing).

Solution: Claude audits codebase against Figma source of truth.

Prompt:

Audit src/components for design system compliance.
Compare against our Figma design system:
https://www.figma.com/design/FILE_KEY

Report:
1. Hardcoded colors not using design tokens
2. Magic number spacing values
3. Components not matching Figma structure
4. Missing responsive patterns

Then propose fixes with token replacements.

Claude output:

## Design System Audit Results

### Issues Found (23 total)

#### Hardcoded Colors (8 instances)
- `src/components/Card.tsx:45` → `#0066CC` should be `var(--interactive-primary)`
- `src/components/Header.tsx:12` → `#F3F4F6` should be `var(--surface-secondary)`
...

#### Magic Numbers (11 instances)
- `src/components/Modal.tsx:23` → `padding: 16px 24px` should be `var(--spacing-4) var(--spacing-6)`
...

#### Structural Mismatches (4 instances)
- `src/components/Button.tsx` → Missing `icon-left` variant present in Figma
...

Workflow C: Token Automation Pipeline

Goal: Figma variable changes automatically propagate to code.

Architecture:

Figma Variables
    ↓ (Tokens Studio export or Figma API)
GitHub Repository (tokens.json)
    ↓ (GitHub Actions CI/CD)
Style Dictionary Transform
    ↓ (generates)
CSS / Tailwind / Platform-specific tokens
    ↓ (commit & deploy)
Production

Setup (one-time):

1. Tokens Studio Plugin: Connect to GitHub repo
2. Style Dictionary Config: Define transform rules
3. GitHub Actions: Auto-run on token updates
4. Claude Role: Review and validate generated tokens

Developer Prompt (after CI runs):

Review the token update from commit abc1234.
Check if any components need updates to consume new tokens.
Generate migration guide if breaking changes exist.

Claude output:

## Token Update Review (v2.3.0 → v2.4.0)

### Changes
- Added: `--spacing-7`, `--spacing-8` (requested by design)
- Changed: `--interactive-secondary` hue shift 5° (brand refresh)
- Deprecated: `--legacy-blue` (remove by Q3)

### Impact Analysis
- 12 components reference `--interactive-secondary` → Auto-updated via token reference
- 3 components use deprecated `--legacy-blue` → Migration needed

### Migration Required
1. `src/components/LegacyButton.tsx:34` → Replace `--legacy-blue` with `--interactive-primary`
2. `src/components/OldCard.tsx:67` → Replace with new token
3. `src/utils/theme.ts:12` → Update theme export

### Migration Script
[Claude generates codemod or find/replace script]

Workflow D: Visual Iteration Loop (Figma + Playwright)

Goal: Automated visual regression testing against Figma designs.

MCP Stack: Figma MCP + Playwright MCP

Setup:

Claude Code accesses:
- Figma (design source of truth)
- Playwright (automated browser testing)

Prompt:

Take a screenshot of our Button component in all variants.
Compare against Figma frames from:
https://www.figma.com/design/FILE_KEY → "Button Tests" page

Report any visual differences (color, spacing, typography).

Workflow:

1. Claude reads Figma frames (expected state)
2. Claude uses Playwright to screenshot live components (actual state)
3. Claude compares (pixel diff or visual inspection)
4. Reports discrepancies with fix suggestions

Example output:

## Visual Regression Report

### ✅ Matching (5/7)
- Button/Primary/Default
- Button/Primary/Hover
- Button/Secondary/Default
...

### ❌ Mismatches (2/7)

#### Button/Primary/Disabled
- **Issue**: Text opacity 0.4 in code, 0.5 in Figma
- **Fix**: Update `disabled:opacity-50` → `disabled:opacity-40`
- **File**: `src/components/Button.tsx:23`

#### Button/Large
- **Issue**: Padding 12px in code, 16px in Figma
- **Fix**: Update `py-3` → `py-4` (16px)
- **File**: `src/components/Button.tsx:18`

Code Connect Setup

Code Connect is Figma's no-code tool for linking design components to code snippets. This enhances Claude's ability to generate correct code.

What It Does

• Designers annotate Figma components with code examples
• Claude reads these annotations via MCP
• Generated code matches team conventions automatically

Setup (for Designers)

1. In Figma Dev Mode → Select component → Code Connect panel
2. Add code snippet showing how component is used:

// Example Code Connect annotation in Figma
<Button variant="primary" size="lg">
  Click me
</Button>

3. Claude sees this when asked to implement, uses team's exact patterns

Benefits

Reference: Read more at parallelhq.com/blog (Code Connect UI article)

Alternative: Pencil (IDE-Native Canvas)

Overview: Pencil brings infinite design canvas directly into Claude Code/Cursor/VSCode, eliminating external tool switching and enabling design-as-code workflows.

Architecture

Core Innovation: Unlike Figma (cloud-based) or Excalidraw (standalone), Pencil embeds the design canvas directly in your IDE where Claude and your code live.

Traditional Workflow:
Figma (design) → Export → Claude Code → Implementation → Manual sync

Pencil Workflow:
IDE Canvas (design + AI agents + code) → Git commit → Continuous alignment

Key Features:

• WebGL Canvas: Infinite, performant, fully editable
• AI Multiplayer Agents: Parallel agents process design collaboratively
• Git-Native: .pen files (JSON format) version-controlled alongside code
• MCP Bi-Directional: Full read+write access (not just read like Figma MCP)
• Figma Import: Copy-paste directly from Figma preserving vectors and styles

Pencil vs. Figma MCP

When to Use Pencil

✅ Good Fit:

• Team uses Cursor or VSCode + Claude Code as primary environment
• Engineer-designers comfortable with terminal/IDE workflows
• Projects requiring tight design-code alignment (design-as-code paradigm)
• Desire for git-native design versioning (branch protection, rollback, etc.)
• Want to leverage parallel AI agents for design automation

⚠️ Consider Carefully:

• Traditional design team (non-technical) → Figma may be better
• Need enterprise SLA/support → Pencil still maturing
• Complex design system with 50+ components → Figma ecosystem more mature
• Team not using Cursor/VSCode → Limited compatibility

Setup

1. Install Pencil extension:

   • Visit pencil.dev
   • Follow installation for Cursor/VSCode/Claude Code
   • Create account (currently free)

2. Create first canvas:

   # Open IDE, launch Pencil extension
   # Create new .pen file in your repo
   # Design on infinite canvas
   

3. Git workflow:

   git add design/homepage.pen
   git commit -m "feat(design): add homepage hero section"
   git push
   

4. Claude integration:

   • Claude can read .pen files via MCP
   • Prompt: "Implement the Button component from design/components.pen"
   • Claude extracts design specs and generates code

Example Prompt

Read the "Hero Section" from design/homepage.pen.

Implement as React component with:
- Responsive behavior matching canvas breakpoints
- Animations from design (fade-in, slide-up)
- Copy exactly as specified in canvas
- Use Tailwind for styling

Ensure pixel-perfect match with design specs.

Founder & Backing

Tom Krcha (CEO, Pencil):

• Co-founder Adobe XD (2014-2018), 10 years at Adobe
• Prior exits: Alter Avatars (acquired by Google), Around (acquired by Miro)
• 14+ years development experience

Funding: a16z Speedrun (~$1M) + KAYA VC

Traction: 1M+ views on launch, thousands of signups including Microsoft, Shopify, Uber executives.

Maturity Note

⚠️ Status: Launched January 2026 (very recent). Strong early signals but documentation and ecosystem still maturing.

Recommendations:

• Production projects: Pilot with 1-2 non-critical features first
• New projects: Safe to adopt for teams on Cursor/Claude Code
• Traditional workflows: Stick with Figma MCP until Pencil matures (3-6 months)

Monitor: Pricing announcement, public GitHub repo, mature documentation expected Q2 2026.

Example Prompts

Component Implementation

Implement the "Card/Product" component from our Figma design system:
[Figma URL]

Requirements:
- Use our existing design tokens from tailwind.config.ts
- Include hover states matching Figma interactions
- Implement all variants (default, featured, compact)
- Add TypeScript types for all props

Design System Expansion

Our design team added a new "Badge" component to Figma:
[Figma URL → Badge frame]

Generate:
1. React component with all variants
2. Storybook stories
3. Unit tests for prop combinations
4. Update design system docs

Token Validation

Compare the color tokens in our Tailwind config against
Figma variables from: [Figma URL]

Report any mismatches and generate update script.

Responsive Implementation

Implement the "Hero" section from Figma with exact responsive behavior:
[Figma URL → Hero/Responsive frame]

Figma has 3 breakpoints configured. Match these precisely.

Accessibility Audit

Review the "Modal" component implementation against Figma specs:
[Figma URL]

Check:
- Focus management matches Figma's interaction flow
- Color contrast meets WCAG AA (Figma has contrast checker)
- Keyboard navigation (Figma annotations specify tab order)

Design QA Before Handoff

Review the "Checkout Flow" frames for implementation readiness:
[Figma URL → Checkout Flow page]

Check:
- All interactive states defined (hover, focus, disabled, error)
- Variables used consistently (no magic values)
- Auto Layout constraints are implementable
- Missing anything needed for production code?

Multi-Component Atomic Implementation

Implement the atomic design system components in order:

1. Atoms: [Figma URL → Atoms page]
   - Button, Input, Label, Badge

2. Molecules: [Figma URL → Molecules page]
   - FormField (Label + Input + Error)
   - SearchBar (Input + Button)

3. Organisms: [Figma URL → Organisms page]
   - LoginForm (using molecules)

Ensure each level only imports from lower levels.

Team Adoption Patterns

For Product Designers

New workflow:

1. Design in Figma with proper variable structure
2. Use Figma Make for rapid prototyping
3. Export to Figma Design with Dev Mode enabled
4. Share file URL + specific frames with dev team
5. Claude consumes design → Generates implementation
6. Designer reviews code output visually (not reading code)

Key insight: Designers don't need to learn code. They review implementation by visual comparison against Figma.

For Developers

New workflow:

1. Receive Figma URL from designer
2. Prompt Claude to implement from Figma source
3. Review generated code for architecture fit
4. Run visual comparison (Playwright or manual)
5. Commit production-ready component

Time saved: Skip manual pixel-perfect implementation. Focus on logic, not layout matching.

For Product Managers

New capability: Request design implementation estimates based on Figma frames.

Prompt for PMs:

Review the "Dashboard Redesign" Figma file:
[Figma URL]

Estimate implementation complexity:
- How many new components needed?
- Which existing components can be reused?
- Any technical blockers?

Provide rough timeline for dev implementation.

Claude output:

## Implementation Analysis

### Scope
- 12 frames total
- 4 new components (DataTable, MetricCard, FilterPanel, DateRangePicker)
- 8 existing components reused

### Complexity Assessment
- **Low**: MetricCard (similar to existing Card, 1-2h)
- **Medium**: FilterPanel (multi-select logic, 4-6h)
- **High**: DataTable (sorting, pagination, virtualization, 2-3 days)
- **High**: DateRangePicker (third-party library integration, 1-2 days)

### Technical Considerations
- DataTable needs backend API for server-side pagination
- DateRangePicker: evaluate date-fns vs dayjs vs native
- FilterPanel state management (local vs global)

### Estimated Timeline
- Development: 5-7 days
- Code review + QA: 2 days
- Total: 1.5-2 weeks

Anti-Patterns

Implementation Roadmap

Phase 1: Foundation (Week 1-2)

Goal: Basic Figma → Claude → Code pipeline

• Install Figma MCP Server
• Configure personal access token
• Test connection: Claude reads public Figma file
• Create project CLAUDE.md with design system conventions
• Implement 3-5 simple components (Button, Input, Badge)
• Establish visual QA process (manual comparison)

Success criteria:

• Claude generates component from Figma URL
• Output matches design visually
• Dev team understands workflow

Phase 2: Scaling (Week 3-4)

Goal: Full design system implementation + automation

• Implement 20+ components from Figma library
• Set up token automation (Tokens Studio + Style Dictionary)
• Create component testing suite (Storybook + visual regression)
• Train designers on variable hygiene
• Document team conventions in CLAUDE.md
• Run first design system drift audit

Success criteria:

• 80%+ of UI components auto-generated
• Token updates propagate automatically
• Designers confident in handoff process

Phase 3: Orchestration (Week 5+)

Goal: Multi-MCP workflows + continuous sync

• Integrate Playwright MCP for automated visual testing
• Set up CI/CD for design-code parity checks
• Create Figma → GitHub → Production pipeline
• Implement design system governance (linting, audits)
• Enable non-devs to trigger Claude implementations (tickets, Slack)
• Measure metrics (TTM, inconsistency rate, dev time saved)

Success criteria:

• Design updates → Production in <1 day
• Zero manual design transcription
• Measurable team velocity increase

Resources

Official Documentation

• Figma MCP Server: @modelcontextprotocol/server-figma (GitHub)
• Figma Developer Docs: figma.com/developers
• Style Dictionary: amzn.github.io/style-dictionary
• Tokens Studio Plugin: tokens.studio

Case Studies & Tutorials

• builder.io: "Claude Code + Figma MCP Server: AI Design-to-Code Workflow" (January 2026)

  • builder.io/blog/claude-code-figma-mcp-server
  • Production metrics, workflow examples

• Vladimir Siedykh: "Multi-MCP Orchestration with Claude Code"

  • vladimirsiedykh.com/blog/claude-code-mcp-workflow
  • Figma + Playwright + Linear integration

• Parallel HQ: "Automating Design Systems with AI"

  • parallelhq.com/blog/automating-design-systems-with-ai
  • 75 days saved, Code Connect UI guide

• Composio: "How to Use Figma MCP with Claude Code"

  • composio.dev/blog/how-to-use-figma-mcp-with-claude-code
  • Token hierarchy patterns, setup guide

Community Resources

• Figma Community: Search "Design System Tokens" for starter templates
• MCP Registry: mcp.run → Figma server examples
• Discord: Anthropic Discord → #mcp-servers channel

Related Workflows

• Working with Images — Claude Code image analysis
• ASCII Art & Wireframing — Low-fidelity design iteration
• Playwright MCP — Visual regression testing

See Also

• ../ultimate-guide.md#figma-mcp — Main guide Figma MCP section
• examples/claude-md/product-designer.md — Product Designer CLAUDE.md template
• ../cheatsheet.md — Quick reference