Figma Generate Design skill
Figma's skill for building or updating screens in a Figma file from your code, using your design system's components and variables.
Official · Figma| Author | Figma |
|---|---|
| Licence | No licence file; Figma Developer Terms apply |
| Source | figma/mcp-server-guide · SKILL.md on GitHub |
| Project website | developers.figma.com/docs/figma-mcp-server |
| Works in | Claude Code · Codex, Gemini CLI, Cursor, Copilot |
| Read from the source on |
The figma-generate-design skill is Figma's official agent skill for building a screen in Figma from code or a description: a page, modal, drawer, form or dashboard. Working through the Figma MCP server, Claude finds your design system's components, variables and text styles, then assembles the view section by section from real instances instead of drawing boxes with hard-coded colours.
The result is meant to be editable Figma layers, never a pasted screenshot: the skill forbids importing a flattened picture of the UI as the deliverable. It needs a Figma account, the Figma MCP server (remote) and a target file, and it works together with Figma's figma-use skill, which holds the rules for every canvas script.
From the skill's SKILL.md: Code-to-design output must recreate the UI as editable text, components, icons, and hierarchy.
What it does
- Reads your source code (or description) and lists the view's sections and the UI components each one uses, and finds the product's real font instead of defaulting to Inter.
- Looks up those components through Code Connect files in your codebase, searches your Figma libraries, and inventories existing screens for component keys, variables and text and effect styles.
- Creates one wrapper frame away from existing content and builds every section inside it, importing design-system instances and binding variables rather than raw values.
- Turns repeated elements into local components with instances, and imports icons as SVG.
- Checks the result: node counts, one composition screenshot compared with the source, and an explicit check that text uses the product font; then fixes only what is wrong.
- For a web app with photos or logos, runs Figma's page capture in parallel and copies its images into the editable build, then deletes the capture.
What you get
- A new or updated frame in your Figma file, made of design-system component instances, local components, variables and styles.
- No file on your computer: the work lands in Figma.
When to use it
- Your app's screens exist in code and you want them in Figma, built from your published components.
- A screen in Figma is out of date and should match what is now in code.
- You want a new screen, modal or dashboard mocked up in Figma using the design system you already have.
When not to use it
- The deliverable is a reusable component or token set for your library: Figma points to its figma-generate-library skill for that.
- You want code from a Figma design (the opposite direction): that is what figma-design-to-code and the MCP server's design-context tools are for.
- Your file has no published design system or team library: the skill expects components to reuse, so results without one are weaker.
- You cannot connect the remote Figma MCP server (for example your organisation blocks it): the skill cannot work without it.
What it needs
- A Figma account with access to the target file. Writing outside your drafts needs a Full seat; Figma says writing to the canvas is free during the beta and will later be a usage-based paid feature.
- The remote Figma MCP server (https://mcp.figma.com/mcp) connected to your agent and signed in to Figma.
- Figma's figma-use skill as well: figma-generate-design requires it before any canvas script. The Figma plugin installs both.
- A target Figma file (URL or file key), and ideally a published design system or team library in it.
- The source code or a description of the screen. Read tools count against Figma's rate limits: up to 20 calls a month on the Starter plan or a View or Collab seat; write tools are exempt.
Files in the skill
| File or folder | What it is for |
|---|---|
SKILL.md | The required six-step workflow (understand, discover components/variables/styles, wrapper frame, build sections, validate, update), its boundaries and error recovery. |
references/componentization.md | Why and how to build repeated elements as components with instances on the first pass. |
references/discover-product-font.md | Where to find the product's real font in code and how to verify it after building. |
Licence as found in the repository on 2026-11-14: No licence file; Figma Developer Terms apply.
Example prompts
Type one of these after installing; naming the skill makes sure it is used.
- Use figma-generate-design to build our pricing page (src/pages/pricing.tsx) in this Figma file: <file URL>. Use components from our Acme UI library.
- Update the Settings modal frame in Figma to match the current SettingsModal.tsx, swapping any components that changed.
- Create a dashboard screen in Figma with a header, three stat cards and a table of recent orders, using our design system variables for colours and spacing.
- Take the running app at localhost:3000/checkout and build it in Figma as editable layers; capture it first so the product images come across.
How to install figma-generate-design
Step 1: connect the Figma MCP server
In Claude Code the easiest route is Figma's plugin from Anthropic's official marketplace: it adds the MCP server settings and all of Figma's skills, including figma-use and figma-generate-design. Run the command in your shell, start Claude Code, then run /mcp and follow the browser steps to sign in to Figma.
Without the plugin, add the server by hand and install the two skill folders yourself (skills-folder steps below; copy figma-use as well).
claude plugin install figma@claude-plugins-official
# or, by hand:
claude mcp add --transport http figma https://mcp.figma.com/mcpIn Cursor, Gemini CLI and VS Code
Figma's guide installs its plugin in Cursor with the /add-plugin figma command in agent chat, and as an extension in Gemini CLI, followed by /mcp auth figma inside Gemini CLI. In VS Code, add an HTTP MCP server with the same URL (Command Palette, MCP: Add Server) and use Copilot's agent mode.
/add-plugin figma
gemini extensions install https://github.com/figma/mcp-server-guideNot for claude.ai uploads or the Claude API Skills endpoint
The skill depends on live MCP tools (use_figma, search_design_system and others). The Claude API skills sandbox has no network access, and we have not verified a claude.ai setup for it, so this page gives only the agent setups above.
A skill is a folder with a SKILL.md file. You install it once; the agent reads its short description at the start of each session and loads the full instructions only when a request matches. Pick the place you use:
Claude Code: from the plugin marketplace
- Start Claude Code in a terminal with
claude. - This plugin is in Anthropic's official marketplace, which Claude Code adds for you the first time you start it.
- Install the plugin. Claude Code opens its details first: check what it will add (skills, hooks, MCP servers), then choose whether it is for you, for everyone on this repository, or for you in this repository only.
/plugin install figma@claude-plugins-official - Type / and look for
/figma:figma-generate-design, or simply ask for the task.
From your shell instead of a session:
claude plugin install figma@claude-plugins-officialThe plugin installs as a whole, so other skills come with it, including figma-3d-transforms, figma-code-connect, figma-code-workflows, figma-create-new-file, figma-design-to-code, figma-generate-diagram, figma-generate-library, figma-generative-plugins, figma-implement-motion, figma-shaders, figma-spec, figma-swiftui, figma-use, figma-use-figjam, figma-use-motion, figma-use-slides.
Claude Code: copy the folder into your skills folder
Put the skill's folder in ~/.claude/skills/ to use it in every project on this computer, or in .claude/skills/ inside a repository to share it with everyone who works there (commit it).
git clone --depth 1 https://github.com/figma/mcp-server-guide.git
mkdir -p ~/.claude/skills
cp -R mcp-server-guide/skills/figma-generate-design ~/.claude/skills/
cp -R mcp-server-guide/skills/figma-use ~/.claude/skills/Claude Code notices new or changed skills in these folders during the session. If the skills folder did not exist when the session started, run /reload-skills once. Then type /figma-generate-design or just describe the task.
On Windows, run the commands in Git Bash or WSL, or copy the folder into the same place with File Explorer.
Codex, Gemini CLI, Cursor and GitHub Copilot
These agents read the same SKILL.md folders. The one folder all four accept inside a project is .agents/skills/, so this copies the skill there:
git clone --depth 1 https://github.com/figma/mcp-server-guide.git
mkdir -p .agents/skills
cp -R mcp-server-guide/skills/figma-generate-design .agents/skills/
cp -R mcp-server-guide/skills/figma-use .agents/skills/| Agent | Project folder | Your user folder | Notes |
|---|---|---|---|
| OpenAI Codex | .agents/skills/ | ~/.agents/skills/ | Type $ or run /skills to pick a skill; Codex also chooses one when your task matches. |
| Gemini CLI | .agents/skills/.gemini/skills/ | ~/.agents/skills/~/.gemini/skills/ | Or install from Git with gemini skills install <git URL> --path <folder> --consent. Google replaced Gemini CLI with Antigravity CLI for free-tier and Google One users on 18 June 2026; Antigravity uses .agents/skills/ in a project and ~/.gemini/config/skills/ for you. |
| Cursor | .agents/skills/.cursor/skills/ | ~/.agents/skills/~/.cursor/skills/ | Type / in Agent chat to pick a skill. Cursor also reads Claude Code's .claude/skills/ folders. |
| GitHub Copilot | .github/skills/.agents/skills/.claude/skills/ | ~/.copilot/skills/~/.agents/skills/ | Works in Copilot CLI, the cloud agent, and agent mode in VS Code, Visual Studio and JetBrains (IDE support depends on the version). Or run gh skill install OWNER/REPOSITORY SKILL (public preview, GitHub CLI 2.90.0 or later). |
Sources for these steps
Checked against each vendor's documentation on 2026-11-14:
How it builds a screen
The skill is strict about order. It discovers before it builds: Code Connect files in your repository are checked first, then your libraries are searched (one query per intent, with synonyms such as "pill", "tab" and "chip"), then existing screens are inventoried. Only gaps that the design system does not cover are built by hand.
Everything is built inside one wrapper frame created up front, because moving nodes between canvas scripts later fails silently. Work runs in retry-safe phases, so an error affects only the current phase and earlier sections stay intact.
Validation ends with a single composition screenshot compared with the source and a font-family check, since a script can load the wrong font without any error.
Images and AI credits
The canvas scripts cannot fetch images from URLs. For web apps with photos, avatars or logos, the skill captures the running page into the same file with Figma's capture tool and copies the image fills across. If no image is available, it can generate placeholders with Figma AI, and says so, because that uses Figma AI credits.
Questions about figma-generate-design
Is the figma-generate-design skill free?
The skill files are free to read and install, but they are not open source: the repository has no licence file and Figma's Developer Terms apply. Using it needs a Figma account. Figma says writing to the canvas is free during the beta and will later become a usage-based paid feature, and read tools are rate-limited by plan and seat.
Do I need the Figma MCP server?
Yes. Every step runs through the remote Figma MCP server's tools, such as use_figma to build on the canvas and search_design_system to find components. Without the server connected and signed in, the skill has nothing to work with.
Which Figma plan or seat do I need?
Figma's developer documentation says writing outside your drafts needs a Full seat. Read tools are limited per plan and seat: up to 20 calls a month on the Starter plan or a View or Collab seat, and per-day and per-minute limits on Professional, Organization and Enterprise Dev or Full seats. Tools that write to files are exempt.
Will I get editable layers or a screenshot?
Editable layers. The skill's hard rule is that the output must recreate the UI as editable text, components, icons and hierarchy; a screenshot may be used only as a visual reference and is deleted afterwards.
What if my Figma file has no design system?
The skill lists a published design system or access to a team library as a prerequisite. Without one it has no components or variables to reuse; for building the library itself, Figma provides the figma-generate-library skill.
Does it work in Cursor, Codex or Gemini CLI?
Figma's guide gives setups for Claude Code, Cursor, Gemini CLI and VS Code with Copilot, and says other tools that support Streamable HTTP MCP servers can connect too. The skill folder itself follows the SKILL.md standard, so any agent that reads skills folders and has the Figma MCP server connected can load it.
More AI skills
Related skills
Figma Use skill (use_figma)
Official · FigmaThe base skill behind every Figma write: the rules an agent follows when it runs Plugin API scripts in your file through use_figma.Figma Design to Code skill
Official · FigmaFigma's skill for turning a Figma frame into code in your project, reusing your components and assets instead of pasting generated markup.Figma Generate Library skill
Official · FigmaFigma's skill for building a design system in Figma from your code: variables, light and dark modes, styles and components with variants.Brand Guidelines skill
Official · AnthropicApplies Anthropic's brand colours and fonts to slides and documents. Best used as a template for a brand skill of your own.Thumbnail Design skill
Community · Frank HeijdenrijkBrief honest, mobile-readable video thumbnails and plan a native comparison test.Design and Templates skill
Community · Frank HeijdenrijkSpecify a reusable brand kit, template family and production workflow for recurring social graphics.