Figma for Agents (2026): use_figma MCP for Claude Code

Overview

Figma for Agents opens the Figma canvas to AI coding agents through a new Model Context Protocol (MCP) tool called use_figma. Announced on March 24, 2026 and currently in open beta, it lets clients such as Claude Code, Cursor, Copilot in VS Code, Codex, and Augment move beyond read-only file inspection and actually create frames, assemble screens, apply variables, and build components linked to an existing design system.

The feature is aimed at product design teams and design-engineering partnerships that already maintain a Figma library — the kind of teams typically comparing AI UI design tools and want to delegate repetitive canvas work — generating variant sets, restructuring hardcoded layouts into token-driven ones, or turning code specs into components — to an agent without leaving their MCP client. It pairs the tool with "Skills," markdown instruction files that teach an agent how a specific team's library should be used.

Unlike earlier AI design tools that generated isolated suggestions, Figma for Agents treats the library as machine-readable context, so output lands in real files with real components rather than detached images. The beta is free during the preview period and will later become a usage-based paid feature tied to existing Figma seats.

Key Features

Write access to the canvas — Agents can create frames, assemble screens, generate component sets, and edit existing layers, so design work can be delegated instead of only previewed.
Design-system awareness — The use_figma tool resolves library components and variables, helping agents reuse tokens and real components instead of pasting hardcoded hex values or ad-hoc layers.
Skills as markdown instructions — Teams ship reusable .md skill files (nine launch-day community skills plus custom ones) that define library conventions, naming rules, and workflows, giving agents consistent behavior across runs.
Multi-client MCP support — Works with any MCP-compatible client, including Claude Code, Cursor, Copilot in VS Code, Codex, and Augment, so teams don't have to adopt a new editor to use it.
Seat-aware permissions — Full seat holders get write access to shared files, while Dev seats remain scoped to drafts, keeping the existing Figma access model intact when agents act on behalf of a user.
Code-to-design round trips — Agents can take a code spec or Storybook-style description and render it back into linked Figma components, closing a loop that previously required manual handoff between designers and engineers.

How to Get Started

Setup takes a few minutes in any MCP-compatible client:

Enable the beta on your Figma account — Sign in with a Full or Dev seat on a Professional, Organization, or Enterprise plan and opt into the Figma for Agents open beta from account settings.
Install use_figma in your MCP client — In Claude Code, Cursor, Copilot in VS Code, Codex, or Augment, add the Figma MCP server entry pointing at use_figma and authenticate via the Figma OAuth prompt.
Point the agent at a file — Share a Figma file URL in your prompt; the agent will read the file, list available components and variables, and confirm write access before making changes.
Install or write Skills — Browse the launch-day community Skills or drop team-specific markdown files into your project so the agent follows your naming, variant, and token conventions.
Review changes in Figma — Every write action lands in version history, so you can inspect, revert, or branch the agent's output just like any manual edit.

Pricing & Plans

Figma for Agents rides on top of standard Figma plans. The use_figma tool itself is free during the open beta; Figma has stated it will become a usage-based add-on once it exits beta, though final pricing has not been published.

Figma Plan	Full Seat	Dev Seat	Agent Access
Starter	Free (limited files)	—	Read-only via MCP in draft files
Professional	$16/seat/month	$12/seat/month	Full write access during beta
Organization	$55/seat/month (annual)	$25/seat/month	Full write access + admin controls
Enterprise	$90/seat/month	$35/seat/month	Full write access + SSO/SCIM governance

Notes:

Full seats can create and edit in any file the user has access to; Dev seats are limited to draft files when acting through an agent.
Organization and Enterprise plans are annual billing only. Collab seats ($5/seat/month on Organization) remain view-and-comment and cannot trigger agent writes.
Pricing reflects the tier structure that took effect on March 11, 2025; Figma has not announced a separate SKU for agent usage yet.

Best For

Product design teams on Professional or Organization plans with a mature component library they want agents to actually use.
Design-engineering pairs exploring AI UX design workflows who want to generate and update Figma screens from code specs.
Design system maintainers who want to encode library conventions as Skills instead of relying on Confluence docs.
Enterprises with SSO/SCIM needs that want agent write access governed by existing Figma seat permissions.
Teams already running MCP-based workflows and looking to bring the design surface into the same automation loop as their repo.

FAQ

What is the use_figma MCP tool?

use_figma is a Model Context Protocol tool that Figma released in open beta on March 24, 2026. It gives MCP-compatible clients permission to read and write on the Figma canvas — creating frames, components, variants, and applying variables — on behalf of an authenticated user. Before this release, most Figma MCP integrations were read-only, so agents could describe a file but not edit it.

Which AI clients can use Figma for Agents?

Figma officially supports MCP clients including Claude Code, Cursor, Copilot in VS Code, Codex, and Augment. Any client that speaks MCP and can authenticate against Figma should be able to register the use_figma tool. Client-specific setup steps vary, but the underlying protocol and permissions are the same.

What are Skills in Figma for Agents?

Skills are markdown files that tell an agent how to behave on the Figma canvas — for example, "always search the library before generating new components," "use these token names," or "follow this variant structure." Nine community Skills shipped at launch, and teams can author their own. Skills are invoked via slash commands in clients like Claude Code, making design-system rules portable and versionable.

Is Figma for Agents free?

It is free during the open beta period that began in March 2026. Figma has said it will become a usage-based paid feature once the beta ends, but has not published final pricing. You still need a paid Figma seat (Professional, Organization, or Enterprise) to get write access through an agent.

Do I need a Full seat or will a Dev seat work?

Full seats get write access to any file the user can normally edit, including shared libraries. Dev seats can trigger agent writes only inside draft files, matching Figma's existing Dev Mode boundaries. If your engineering team uses Dev seats, they'll still need a Full seat holder to merge agent output into production files.

Can agents accidentally break existing designs?

Every action the agent takes goes through Figma's normal version history, so you can inspect the diff, revert changes, or branch off like any manual edit. That said, agents are non-deterministic: the same prompt can produce different layer structures between runs, so Figma recommends reviewing output before publishing and using Skills to constrain behavior for predictable tasks.

How does this compare to Figma's earlier AI features?

Earlier Figma AI features — like Make Designs or AI rename — generated suggestions inside the product but didn't let external agents edit files. Figma for Agents is different in two ways: it uses the open MCP standard instead of a closed in-app flow, and it gives agents structured write access to real components and variables rather than emitting detached images. It's designed to be the piping your existing agent uses, not a separate UI.

Is my Figma data used to train AI models?

Figma's data and privacy policy governs use_figma the same as other Figma features. Content from Organization and Enterprise plans is not used to train AI models by default, and admins can configure AI settings at the workspace level. For Professional plans, review the current AI content controls in your account settings before granting agent access to sensitive files.

What happens to my agent workflows after the beta ends?

Figma has said the tool will transition to usage-based billing rather than being removed. Existing Skills and MCP client configurations are expected to keep working; teams will need to budget for the new pricing once it's announced. Until then, beta users can rely on free access.