Conduit Design – Agente IA para Figma
conduit-design/conduit_design
Automação agentica de designs no Figma via prompts naturais
Ver no GitHubMCP server que transforma agentes IA em colaboradores nativos do Figma, permitindo automação completa de designs através de prompts em linguagem natural. Executa transformações em lote, gera código production-ready e integra-se com editores como Cursor e VSCode.
Instalação
npx convez-ai conduit-design/conduit_design1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automação de alterações em massa em projetos Figma
- Geração de código production-ready a partir de designs
- Integração de IA generativa no fluxo de design
- Transformação de hierarquias complexas em segundos
Veredito

Ferramenta sólida e multiplataforma que leva automação IA a um nível profissional no Figma. Exige setup técnico e MCP-host compatível, mas entrega valor real para equipes que trabalham com design em escala.
README
A tradução para o português deste README ainda está na fila.
conduit.design
Turn any LLM-powered agent into a first-class Figma collaborator. No more clicking through menus or manual adjustments. Just describe what you want, and watch it happen.
- Features
- Prerequisites
- Quick Start
- Configuration
- Export Capabilities
- Supported Applications
- Tool API
- File Sandboxing
- Privacy & Data Usage
- Auto-Updates
- Troubleshooting
- Limitations
Features
- Full Figma API access — Batch transform complex designs with simple prompts
- Instant AI editing — Transform complex design hierarchies in seconds
- Code Export — Generate production-ready code from design instantly
Prerequisites
Before installation, ensure you have:
- Figma Desktop App (Free version or Paid, browser version of Figma not supported)
- MCP-compatible application (Cursor, VSCode, AntiGravity, or another MCP-enabled editor.)
Quick Start
Step 1: Configure Your MCP Application
Add to your MCP configuration file (~/.cursor/mcp_settings.json for Cursor, .vscode/mcp.json for VSCode, or platform-specific location for others):
{
"mcpServers": {
"conduit": {
"command": "/bin/bash",
"args": ["-c", "curl -sSL https://conduit.design/run.sh | bash"]
}
}
}
Step 2: Install and Import the Figma Plugin
- Start your MCP application (e.g., Cursor or VSCode) — this automatically downloads and unpacks the Conduit Figma plugin to
~/.conduit/figma-plugin/ - Open Figma Desktop → Navigate to
Plugins→Development→Import plugin from manifest… - Select the manifest file:
~/.conduit/figma-plugin/manifest.json - Copy the channel ID displayed in the Conduit plugin UI (e.g.,
"purple-owl-27")
Step 3: Connect & Create
In your AI assistant (e.g., Cursor chat):
User: Talk to Figma on channel "purple-owl-27"
Agent: ✅ Connected to Figma on channel "purple-owl-27"
User: Create a modern navigation bar with our brand color #3366FF
Agent: ✅ Created navigation bar with brand color #3366FF
You're good to go! The AI can now read, modify, and create Figma designs.
Configuration
[!WARNING] Security Notice: Conduit requires file system access. Only add trusted directories to
ALLOWED_ROOTS. Always back up your projects before granting write permissions.
Basic Configuration
To customize Conduit's behavior, add environment variables to your MCP configuration:
{
"mcpServers": {
"conduit": {
"command": "/bin/bash",
"args": ["-c", "curl -sSL https://conduit.design/run.sh | bash"],
"env": {
"CHANNEL_KEY": "purple-owl-26",
"PORT": "3055",
"PROJECT_ROOT": "/Users/John/my-project",
"ALLOWED_ROOTS": "/Users/John/Desktop/experiments|/Users/John/Downloads/project-xyz"
}
}
}
}
Environment Variables Reference
CHANNEL_KEY: A permanent channel ID to pair the MCP server with the Figma plugin. If you don't provide one, it will be auto-generated. You can copy the ID from the Conduit plugin UI inside Figma. (Example:"purple-owl-26")PORT: The WebSocket port the plugin connects to. (Default:3055)PROJECT_ROOT: The directory where Conduit will store its logs and temporary files. Many MCP hosts default this to your current workspace, but some may require you to set it explicitly. (Example:"/Users/John/my-project")ALLOWED_ROOTS: A pipe-delimited (|) list of absolute directory paths where Conduit is allowed to create and overwrite files. Be cautious with broad paths like~/or/. (Example:"/Users/John/my-project|/Users/John/experiments")
[!IMPORTANT]
PROJECT_ROOTdoes not limit whereALLOWED_ROOTScan point—they are independent.- Only directories listed in
ALLOWED_ROOTSallow file overwrites. All other locations are read-only or create-only.
Windows Configuration
On Windows, use Windows-style paths and PowerShell syntax. Example:
{
"mcpServers": {
"conduit": {
"command": "powershell.exe",
"args": [
"-NoProfile",
"-Command",
"iwr https://conduit.design/run.ps1 -UseBasicParsing | iex"
],
"env": {
"PROJECT_ROOT": "C:\\Users\\John\\my-project",
"ALLOWED_ROOTS": "C:\\Users\\John\\my-project|C:\\Users\\John\\Desktop\\experiments"
}
}
}
}
[!NOTE] Use double backslashes (
\\) in JSON for Windows paths. At runtime, these become standard Windows paths likeC:\Users\John\....
Instant AI edit Integration (Optional)
To enable AI-powered instant edits, add your provider API keys and model preferences:
{
"mcpServers": {
"conduit": {
"command": "/bin/bash",
"args": ["-c", "curl -sSL https://conduit.design/run.sh | bash"],
"env": {
"CHANNEL_KEY": "purple-owl-26",
"PORT": "3055",
"PROJECT_ROOT": "/Users/John/my-project",
"ALLOWED_ROOTS": "/Users/John/my-project",
"AI_DEFAULT_PROVIDER": "openai",
"AI_DEFAULT_MODEL": "gpt-5-1",
"OPENAI_API_KEY": "sk-proj-...",
"AI_TEMPERATURE": "0.7",
"AI_MAX_TOKENS": "4096"
}
}
}
}
AI Provider Options
| Provider | Supported Models | Environment Key |
|---|---|---|
| OpenAI | gpt-5-1, gpt-5-1-mini | OPENAI_API_KEY |
| Anthropic | claude-sonnet-4-5, claude-opus-4-5 | ANTHROPIC_API_KEY |
gemini-3.0-pro, gemini-3.0-flash | GEMINI_API_KEY |
AI-Specific Environment Variables
| Variable | Description | Default | Example |
|---|---|---|---|
AI_DEFAULT_PROVIDER | AI provider to use for instant edits. | None | openai |
AI_DEFAULT_MODEL | Model to use for instant edits. | None | gpt-5-1 |
AI_TEMPERATURE | Creativity/randomness (0.0-1.0). | 0.7 | 0.7 |
AI_MAX_TOKENS | Maximum tokens for AI responses. | 4096 | 4096 |
Export Capabilities
Conduit can export production-ready websites from Figma frames using the {page-breakpoint-theme} naming convention:
Export Features
- Responsive Design — Automatic mobile, tablet, and desktop breakpoints
- Theming — Dark and light mode support
- Asset Export — Optimized SVG and image exports
- CSS Tokens — Design system variables extracted automatically
- Complete Files — HTML, CSS, and all assets bundled together
- Google Fonts — Automatic imports and optimization
- Multi-Page — Export entire websites with multiple pages
- Hyperlinks — Navigation and interaction support
Export Workflow
- Name your Figma frames using the pattern:
{page-breakpoint-theme}- Example:
home-desktop-light,about-mobile-dark
- Example:
- Use the
exporttool via your AI assistant - Conduit generates complete HTML/CSS/JSX with all assets
Supported Applications
Other MCP-compatible applications can also use Conduit. Below is a list of tested applications:
| Application | Configuration Location | Status |
|---|---|---|
| Cursor | ~/.cursor/mcp_settings.json | ✅ Stable |
| Google Antigravity | Extension settings | ✅ Stable |
| VSCode | .vscode/mcp.json | ✅ Stable |
Tool API
Conduit exposes the following tools to LLM agents. Most tools support both single and batch operations.
Core Design Tools
| Tool | Description |
|---|---|
annotation | Get, set, and delete annotations on Figma nodes. |
asset | Get or set images, SVG, and vector assets on nodes. |
autolayout | Configure auto layout stacks and grids with spacing, padding, and alignment. |
boolean | Perform boolean operations (union, subtract, intersect, exclude) or flatten nodes. |
component | Create, manage, and modify components, instances, and component sets. |
constraint | Set layout constraints for responsive behavior (left, right, top, bottom, center, scale, stretch). |
effect | Apply visual effects like shadows and blurs to nodes. |
grid | Create, update, delete, and retrieve layout grids on frames and components. |
group | Group or ungroup nodes. |
guide | Add, delete, or retrieve ruler guides on pages. |
node | Insert, move, clone, rename, and delete nodes with full hierarchy control. |
page | Create, delete, rename, duplicate, or navigate between pages. |
property | Set node properties like visibility, locked state, and opacity. |
shape | Create shapes (rectangle, ellipse, star, polygon, line, frame) with styling. |
style | Create, update, remove, and apply fill, stroke, text, and effect styles. |
text | Create and style text nodes with typography, alignment, and formatting. |
transform | Set position, size, rotation, or apply transformation matrices. |
variant | Manage component variants and their properties within component sets. |
variable | Create and manage design tokens (colors, numbers, strings, booleans). |
Utility Tools
| Tool | Description |
|---|---|
bulk | Execute multiple MCP commands in sequence with structured results. |
compare | Compare JSON data and generate diffs for tracking changes. |
describe | Get detailed metadata and documentation for any MCP tool. |
info | Retrieve detailed node information in multiple formats with depth control. |
join | Connect to a Figma channel (required before other operations). |
selection | Get or set the current node selection in Figma. |
export | Export Figma frames to production-ready HTML/CSS with responsive breakpoints and themes. |
jsx | Get, set, or AI-edit JSX representations of Figma nodes. |
File Sandboxing:
"Working directory" is what your AI agent app (Cursor / VSCode) advertises as allowed roots. Please avoid adding low-level directories to config, AIs might in rare cases overwrite files unintentionally. Ensure you backup your project files frequently.
- Read Access — Conduit can read files from your entire system.
- Soft-Write Access — Conduit can create new files and folders anywhere on your system but cannot overwrite existing files.
- Safe-Write Access — Conduit can overwrite existing files only within directories listed in
ALLOWED_ROOTS.
Security Best Practices
[!WARNING] Only add trusted directories to
ALLOWED_ROOTS. Avoid broad paths like~/or/that grant write access to sensitive system or user files.
- Limit
ALLOWED_ROOTSto project-specific directories (e.g.,/Users/John/my-project). - Back up your projects regularly before granting write permissions.
- Review directories in
ALLOWED_ROOTSperiodically to ensure they're still appropriate.
Privacy & Data Usage
Only data you actively work on is exposed to external services. See below regarding what gets exposed.
- Conduit is an entirely local application and has no analytics and does not call any external service or server.
- Conduit uses (Cursor / VSCode) as your daily driver. If you trust them you're good to go. Their EULA applies.
- Conduit connects to google-gemini, Anthropic, openAI APIs if you use the "instant edit" feature. Their EULA applies.
Auto-Updates
Conduit automatically updates both the MCP server and the Figma plugin when you restart your MCP application or reload your MCP configuration.
Update Process
- On MCP Restart: Conduit checks for updates to the server binary and Figma plugin.
- Auto-Download: If updates are available, they are downloaded and applied automatically.
- Plugin Sync: The Figma plugin is updated in
~/.conduit/figma-plugin/without requiring manual re-import.
Troubleshooting
Figma Plugin Not Connecting
Issue: The Figma plugin shows "Disconnected" or fails to pair with the MCP server.
Solution:
- Verify the
CHANNEL_KEYin your MCP configuration matches the channel ID shown in the Figma plugin UI. (This step is optional, you can also just give channel_key to the AI when joining) - Ensure the
PORTis not blocked by a firewall or already in use. - Restart both your MCP application and the Figma desktop app.
File Write Errors
Issue: Conduit fails to write or overwrite files.
Solution:
- Check that the target directory is listed in
ALLOWED_ROOTS. - Verify the path uses absolute paths (e.g.,
/Users/John/my-project, not~/my-project). - Ensure you have file system permissions for the target directory.
Plugin Import Issues
Issue: Cannot import the Figma plugin via Plugins → Development → Import plugin from manifest….
Solution:
- Ensure you're using Figma Desktop — Plugin import is not available in the browser version.
- Verify the manifest file exists at
~/.conduit/figma-plugin/manifest.json. - If the file is missing, restart your MCP application to trigger the auto-download.
AI Instant Edit Not Working
Issue: AI-powered instant edits fail or are unavailable.
Solution:
- Verify you've added the appropriate API key (e.g.,
OPENAI_API_KEY,ANTHROPIC_API_KEY,GEMINI_API_KEY) to your MCP configuration. - Check that
AI_DEFAULT_PROVIDERandAI_DEFAULT_MODELare set correctly. - Ensure your API key is valid and has sufficient quota.
Limitations
[!NOTE] Early Access Notice: Conduit is under active development. The following limitations are being actively addressed:
- Instant AI Edit: Batch-mode not yet available (workaround: target parent nodes directly).
- Node Type Coverage: Instant AI Edit supports ~95% of node types; some complex types require turn-by-turn mode.
- Multi-level Styles: Instant AI Edit doesn't support multi-level styles yet; use turn-by-turn mode for these cases.
- Export Formats: Currently limited to HTML, CSS, and JSX. Tailwind, Vue, and other frameworks coming in future releases.
- Bulk Operations: Chaining mixed tool calls in bulk mode is experimental; only select tools currently support this.
© 2025 conduit.design - All rights reserved.