Prototype → Figma Skill
Skill by ara.so — Design Skills collection.
Converts working Claude Code prototypes into structured Figma design files by exploding interaction flows into state-by-state frames, mapping components to your Figma design system, and annotating triggers, transitions, and edge cases natively in Figma.
What This Skill Does
When a user builds a prototype in Claude Code, this skill:
- Analyzes the prototype source — inventories components, maps interaction flows, identifies all UI states
- Maps to design system — searches linked Figma libraries for matching DS components; flags unmatched elements with "No DS match" badges
- Creates one frame per interaction state — every meaningful step in a user flow becomes a separate Figma frame
- Annotates interactions — adds native Figma Dev Mode annotations with filterable categories (Interaction, Navigation, Validation, Error Handling, etc.)
- Adds flow arrows — visual connectors showing user path through states
- Creates overview frame — table of contents with legend and open questions for reviewers
- Optionally links Code Connect — maps Figma components back to codebase
Key Capabilities
- Design system aware — uses real components from target file's linked libraries
- Primitive fallback — builds missing components from Figma primitives (never skips elements)
- Multi-tier output — builds directly in Figma on supported clients, generates spec documents on others
- Code Connect integration — creates bidirectional links between Figma and code
- Automatic file creation — creates new Figma file if no URL provided
When to Use This Skill
User says things like:
- "Take this prototype and put it in Figma so the team can review it"
- "Explode this into Figma frames for async feedback"
- "Create Figma specs from my prototype for design review"
- "Make this prototype reviewable by the design team"
Required MCP Tools
This skill requires the Figma MCP server with these tools available:
use_figma— select target filesearch_design_system— find matching componentsget_design_context— understand file structureget_metadata— read component detailsget_screenshot— verify outputget_code_connect_map— read existing mappingsget_context_for_code_connect— understand component contextget_code_connect_suggestions— get AI suggestions for mappingssend_code_connect_mappings— batch create mappingsadd_code_connect_map— create individual mappingwhoami— get user info for file creationcreate_new_file— create new Figma file
Installation Check
Before using this skill, verify Figma MCP is installed:
# Check Claude Desktop config
cat ~/Library/Application\ Support/Claude/claude_desktop_config.json | grep figma
# Or for Claude Code
cat ~/.claude/mcp_config.json | grep figma
Expected output should include @figma/mcp-server or similar.
Workflow Steps
1. Resolve Target File
If user provides Figma URL:
await use_figma({
url: "https://figma.com/design/abc123/MyFile"
});
If no URL provided, create new file:
// Get user info
const user = await whoami();
// Create new file
const newFile = await create_new_file({
name: `${prototypeName} — Prototype Review`,
team_id: user.team_id // or user.default_team_id
});
await use_figma({ url: newFile.url });
2. Analyze Prototype Source
Read the prototype files and extract:
- Component inventory (Button, Input, Modal, etc.)
- Interaction flows (user actions → state changes)
- UI states (initial, loading, success, error, etc.)
- Data flows (form submissions, API calls, etc.)
Example analysis structure:
interface PrototypeAnalysis {
components: {
name: string; // e.g. "PrimaryButton"
type: string; // e.g. "Button"
props: string[]; // e.g. ["label", "onClick", "disabled"]
variants: string[]; // e.g. ["default", "hover", "disabled"]
}[];
flows: {
name: string; // e.g. "Login Flow"
states: {
id: string; // e.g. "login-initial"
label: string; // e.g. "Login Form (Empty)"
components: string[];
interactions: {
trigger: string; // e.g. "Click 'Sign In'"
action: string; // e.g. "Validate and submit"
nextState: string; // e.g. "login-loading"
condition?: string; // e.g. "If fields valid"
}[];
}[];
}[];
}
3. Map to Design System
For each component in the prototype:
// Search for matching DS component
const results = await search_design_system({
query: componentName,
component_type: componentType, // "Button", "Input", etc.
limit: 5
});
// If match found, get metadata
if (results.length > 0) {
const metadata = await get_metadata({
node_id: results[0].node_id
});
// Store mapping
componentMap.set(componentName, {
figma_node_id: results[0].node_id,
figma_name: results[0].name,
variants: metadata.variants,
props: metadata.properties
});
} else {
// Mark for primitive fallback
componentMap.set(componentName, {
needs_primitive_fallback: true,
primitive_type: componentType
});
}
4. Plan Page Structure
Organize frames by flow:
Page: "Login Flow"
├─ Overview Frame (legend + TOC)
├─ State 1: Login Form (Empty)
├─ State 2: Login Form (Validation Error)
├─ State 3: Login Form (Loading)
├─ State 4: Login Success
└─ State 5: Login Error (Network)
Calculate frame positions:
const FRAME_WIDTH = 1440; // desktop, or 390 for mobile
const FRAME_HEIGHT = 900; // desktop, or 844 for mobile
const HORIZONTAL_SPACING = 200;
const VERTICAL_SPACING = 200;
function calculateFramePosition(index: number, perRow: number = 3) {
const row = Math.floor(index / perRow);
const col = index % perRow;
return {
x: col * (FRAME_WIDTH + HORIZONTAL_SPACING),
y: row * (FRAME_HEIGHT + VERTICAL_SPACING)
};
}
5. Build Frames in Figma
CRITICAL: Never call createComponent() — always use createInstance() for DS components or primitives for fallbacks.
For matched DS components:
// Create instance of DS component
const buttonInstance = await figma.createInstance({
component_id: componentMap.get("PrimaryButton").figma_node_id,
parent_id: frameId,
x: 100,
y: 200,
properties: {
variant: "default",
label: "Sign In"
}
});
For components needing primitive fallback:
// Build from primitives (example: Button)
const buttonGroup = await figma.createFrame({
name: "Button (No DS match)",
parent_id: frameId,
x: 100,
y: 200,
width: 120,
height: 40,
fills: [{ type: "SOLID", color: { r: 0.2, g: 0.4, b: 1 } }],
cornerRadius: 8
});
const buttonLabel = await figma.createText({
parent_id: buttonGroup.id,
characters: "Sign In",
fontSize: 16,
fills: [{ type: "SOLID", color: { r: 1, g: 1, b: 1 } }]
});
// Add "No DS match" badge
const badge = await figma.createFrame({
name: "⚠️ No DS match",
parent_id: buttonGroup.id,
x: -10,
y: -10,
width: 100,
height: 20,
fills: [{ type: "SOLID", color: { r: 1, g: 0.8, b: 0 } }]
});
6. Add Annotations
Use native Figma Dev Mode annotations with categories:
await figma.addAnnotation({
node_id: buttonInstance.id,
category: "Interaction",
text: "Trigger: Click\nAction: Validate form fields\nNext: → Login Form (Loading) if valid, → Login Form (Error) if invalid"
});
await figma.addAnnotation({
node_id: emailInput.id,
category: "Validation",
text: "Required field\nValidation: Must be valid email format\nError message: 'Please enter a valid email'"
});
await figma.addAnnotation({
node_id: errorBanner.id,
category: "Error Handling",
text: "Shown when: Network request fails\nDuration: Persistent until dismissed\nDismiss: Click X icon or retry"
});
Categories to use:
Interaction— clicks, hovers, focusNavigation— route changes, deep linksValidation— input rules, error statesError Handling— error scenarios, recoveryData Flow— API calls, state updatesEdge Cases— empty states, loading, timeouts
7. Add Flow Arrows
Create connector arrows between frames:
await figma.createConnector({
start_node_id: state1Frame.id,
end_node_id: state2Frame.id,
stroke_weight: 2,
stroke_color: { r: 0.5, g: 0.5, b: 0.5 },
label: "Click 'Sign In'"
});
8. Create Overview Frame
First frame in each flow page:
const overviewFrame = await figma.createFrame({
name: "📋 Overview",
width: FRAME_WIDTH,
height: FRAME_HEIGHT,
fills: [{ type: "SOLID", color: { r: 0.98, g: 0.98, b: 0.98 } }]
});
// Add title
await figma.createText({
parent_id: overviewFrame.id,
characters: "Login Flow — Prototype Review",
fontSize: 32,
fontWeight: 700,
y: 40
});
// Add legend
await figma.createText({
parent_id: overviewFrame.id,
characters: `Legend:
• 🟦 Frame = One UI state
• ➡️ Arrow = User action / transition
• 📝 Annotation = Interaction detail
• ⚠️ Badge = No design system match (built from primitives)`,
fontSize: 16,
y: 120
});
// Add TOC
await figma.createText({
parent_id: overviewFrame.id,
characters: `States in this flow:
1. Login Form (Empty)
2. Login Form (Validation Error)
3. Login Form (Loading)
4. Login Success
5. Login Error (Network)`,
fontSize: 16,
y: 250
});
// Add open questions
await figma.createText({
parent_id: overviewFrame.id,
characters: `Open questions for reviewers:
• Should "Forgot Password" link to modal or new page?
• Do we show loading spinner in button or overlay entire form?
• How long should success message persist?`,
fontSize: 16,
y: 450,
fills: [{ type: "SOLID", color: { r: 0.8, g: 0.4, b: 0 } }]
});
9. Create Code Connect Mappings (Optional)
If user wants Code Connect integration:
// Get existing mappings
const existingMaps = await get_code_connect_map({
file_key: fileKey
});
// Get suggestions for unmapped components
const suggestions = await get_code_connect_suggestions({
component_id: buttonComponentId,
code_snippet: `<Button variant="primary" In</Button>`
});
// Create mappings
const mappings = componentMap.entries().map(([codeName, figmaInfo]) => ({
figma_node_id: figmaInfo.figma_node_id,
code_location: `src/components/${codeName}.tsx`,
code_snippet: getComponentSnippet(codeName),
framework: "react" // or "vue", "svelte", etc.
}));
await send_code_connect_mappings({
mappings: mappings
});
10. Verify and Present
// Take screenshot of output
const screenshot = await get_screenshot({
node_id: pageId,
scale: 0.5 // overview scale
});
// Present to user
return `✅ Prototype converted to Figma!
File: ${figmaFileUrl}
Created:
• ${flowCount} flows
• ${frameCount} state frames
• ${annotationCount} interaction annotations
• ${componentMatchCount} design system components used
• ${primitiveFallbackCount} components built from primitives (flagged with ⚠️)
${codeConnectEnabled ? `Code Connect: ${mappingCount} mappings created` : ''}
Open questions for reviewers:
${openQuestions.join('\n')}
[Screenshot attached]`;
Common Patterns
Pattern: Detecting Prototype Framework
function detectFramework(prototypeFiles: string[]): string {
const hasReact = prototypeFiles.some(f =>
f.includes('import React') || f.includes('from "react"')
);
const hasVue = prototypeFiles.some(f =>
f.includes('<template>') || f.includes('from "vue"')
);
const hasSvelte = prototypeFiles.some(f =>
f.includes('<script>') && f.endsWith('.svelte')
);
if (hasReact) return 'react';
if (hasVue) return 'vue';
if (hasSvelte) return 'svelte';
return 'html'; // vanilla
}
Pattern: Grouping Micro-Interactions
Group very small interactions (hover, focus) into a single frame with annotations:
function shouldGroupInteractions(interactions: Interaction[]): boolean {
// If all interactions are micro-interactions (hover, focus, blur)
const microTypes = ['hover', 'focus', 'blur', 'mouseover', 'mouseout'];
return interactions.every(i => microTypes.includes(i.type));
}
// Instead of 5 frames for hover states, create 1 frame with annotations:
await figma.addAnnotation({
node_id: button.id,
category: "Interaction",
text: `Hover: Background darkens to #1a3d8f
Focus: 2px blue outline appears
Click: Ripple animation, navigate to /dashboard`
});
Pattern: Handling Responsive Variants
function explodeResponsiveStates(state: UIState): UIState[] {
if (!state.hasResponsiveVariants) return [state];
return [
{ ...state, viewport: 'mobile', frameWidth: 390, frameHeight: 844 },
{ ...state, viewport: 'tablet', frameWidth: 768, frameHeight: 1024 },
{ ...state, viewport: 'desktop', frameWidth: 1440, frameHeight: 900 }
];
}
Pattern: Primitive Button Fallback
async function createPrimitiveButton(
parent_id: string,
x: number,
y: number,
label: string,
variant: 'primary' | 'secondary' = 'primary'
) {
const colors = {
primary: { bg: { r: 0.2, g: 0.4, b: 1 }, fg: { r: 1, g: 1, b: 1 } },
secondary: { bg: { r: 0.95, g: 0.95, b: 0.95 }, fg: { r: 0.2, g: 0.2, b: 0.2 } }
};
const group = await figma.createFrame({
name: `Button: ${label} (No DS match)`,
parent_id,
x,
y,
width: Math.max(120, label.length * 10),
height: 40,
fills: [{ type: "SOLID", color: colors[variant].bg }],
cornerRadius: 8,
paddingLeft: 16,
paddingRight: 16
});
await figma.createText({
parent_id: group.id,
characters: label,
fontSize: 16,
fontWeight: 500,
fills: [{ type: "SOLID", color: colors[variant].fg }],
textAlignHorizontal: "CENTER",
textAlignVertical: "CENTER"
});
// Add badge
await figma.createFrame({
name: "⚠️ No DS match",
parent_id: group.id,
x: -12,
y: -12,
width: 110,
height: 24,
fills: [{ type: "SOLID", color: { r: 1, g: 0.8, b: 0 } }],
cornerRadius: 4
});
return group;
}
Troubleshooting
"No Figma MCP tools available"
Cause: Figma MCP server not installed or not running.
Solution:
# Install Figma MCP
npx @figma/mcp-server init
# Or check Claude Desktop config
cat ~/Library/Application\ Support/Claude/claude_desktop_config.json
"Cannot create component"
Cause: Skill is calling createComponent() instead of using instances or primitives.
Solution: NEVER call createComponent(). Always:
- Use
createInstance()for DS components - Build from primitives (
createFrame,createText, etc.) for fallbacks
"Design system search returns no results"
Cause: Target file has no linked libraries, or component naming mismatch.
Solution:
- Verify linked libraries in Figma file settings
- Try fuzzy search:
search_design_system({ query: "btn primary" })instead of exact "PrimaryButton" - Fall back to primitives if no match after 2-3 search attempts
"Too many frames created"
Cause: Exploding every micro-interaction into separate frames.
Solution: Group hover/focus/blur into single frame with annotations (see pattern above).
"Annotations not visible"
Cause: User not in Dev Mode, or annotations have no category.
Solution: Always specify category. Remind user to switch to Dev Mode in Figma.
"Code Connect mappings fail"
Cause: Invalid code snippet format, or component not in file.
Solution:
// Get context first
const context = await get_context_for_code_connect({
component_id: nodeId
});
// Then create mapping with valid snippet
await add_code_connect_map({
figma_node_id: nodeId,
code_location: "src/components/Button.tsx",
code_snippet: `import { Button } from "./Button";\n\n<Button variant="primary">Click me</Button>`,
framework: "react"
});
Output Tier Compatibility
This skill produces different outputs based on client capabilities:
Tier 1 (Full build in Figma): Claude Code, Claude Desktop, Cursor, VS Code, Copilot CLI, Augment Code, Factory, Firebender, Codex
- Builds frames directly in Figma
- Creates Code Connect mappings
- Adds native annotations
Tier 2 (Spec document): Android Studio, Gemini CLI, Kiro, Amazon Q, Openhands
- Generates markdown spec document
- Includes frame-by-frame descriptions
- No direct Figma manipulation
Tier 3 (Build only): Replit
- Builds frames in Figma
- No Code Connect support
Check available tools at runtime:
const hasCodeConnect = tools.includes('get_code_connect_suggestions');
const canBuildInFigma = tools.includes('use_figma');
Environment Variables
This skill does not require API keys — it uses Figma MCP tools which handle authentication via OAuth.
Users must have Figma account connected to their Claude client.
Best Practices
- Always verify file access — call
use_figma()before any other Figma operations - Never skip elements — if no DS match, build from primitives and flag with badge
- Group micro-interactions — don't create 10 frames for hover states
- Organize by flow — one page per user flow, not one giant page
- Use consistent naming — "State 1: [Description]" format for frame names
- Add legend on every page — overview frame explaining notation
- Flag open questions — surface design decisions that need input
- Take screenshots — verify output before presenting to user
- Provide file URL — always share direct link to Figma file in response
Example Full Workflow
// User: "Put this login prototype in Figma"
// 1. Create new file (no URL provided)
const user = await whoami();
const file = await create_new_file({
name: "Login Prototype — Review",
team_id: user.team_id
});
await use_figma({ url: file.url });
// 2. Analyze prototype
const analysis = analyzePrototype(prototypeFiles);
// → { flows: [{ name: "Login Flow", states: [...] }], components: [...] }
// 3. Map components
for (const comp of analysis.components) {
const results = await search_design_system({
query: comp.name,
component_type: comp.type
});
if (results.length > 0) {
componentMap.set(comp.name, results[0]);
} else {
componentMap.set(comp.name, { primitive: true, type: comp.type });
}
}
// 4. Create page
const page = await figma.createPage({ name: "Login Flow" });
// 5. Create overview frame
const overview = await createOverviewFrame(page.id, analysis.flows[0]);
// 6. Create state frames
for (let i = 0; i < analysis.flows[0].states.length; i++) {
const state = analysis.flows[0].states[i];
const pos = calculateFramePosition(i + 1); // +1 to skip overview
const frame = await figma.createFrame({
name: `State ${i + 1}: ${state.label}`,
parent_id: page.id,
x: pos.x,
y: pos.y,
width: 1440,
height: 900
});
// Build components
for (const compName of state.components) {
const mapping = componentMap.get(compName);
if (mapping.primitive) {
await createPrimitiveButton(frame.id, 100, 100, compName);
} else {
await figma.createInstance({
component_id: mapping.figma_node_id,
parent_id: frame.id
});
}
}
// Add annotations
for (const interaction of state.interactions) {
await figma.addAnnotation({
node_id: frame.id,
category: "Interaction",
text: `${interaction.trigger} → ${interaction.action}`
});
}
// Add flow arrow to next state
if (i < analysis.flows[0].states.length - 1) {
await figma.createConnector({
start_node_id: frame.id,
end_node_id: nextFrameId
});
}
}
// 7. Screenshot and present
const screenshot = await get_screenshot({ node_id: page.id });
return `✅ Prototype in Figma: ${file.url}`;
This skill is part of the Design Skills collection at ara.so. For issues or contributions, see the GitHub repo.