Figma Agent
Figma · Figma Agent: AI Design Agent Built Into the Figma Design Canvas
Figma Agent is the AI design agent built into Figma Design. It generates and remixes screens with your team's real components, styles and variables, makes bulk edits across a file, and turns comment threads into design changes. It suits product designers and design-system teams who already work in Figma and pay for usage with Figma AI credits.
Best for
Product designers and design-system teams already working in Figma Design who want an agent that edits their real files with their own components and variables, for exploration, bulk edits, documentation and acting on feedback
Not ideal for
People who do not design in Figma, teams that need working code or interactive prototypes from a prompt (Figma Make or an app builder fits better), and organizations that need a fixed, predictable cost per task or an agent they can call from their own systems
Who it's for
Product designers, design-system maintainers and product teams who build interfaces in Figma Design
Capabilities
- Generates new screens and layouts from a prompt, and remixes an existing design into several stylistic directions on the same canvas
- Uses a connected library's components, styles and variables, with @-mentions to point it at specific components, tokens or variables
- Library guideline files (.md, .txt or .json, up to 200 KB per library, on paid plans) that are passed to the agent automatically when someone designs with that library
- Bulk edits: update typography across a file, set many component instances to a variant, change spacing across a flow, swap placeholder copy and images, or convert screens to dark mode
- Edits auto layout, component properties and variants, styles, variables and modes, and can update main components and design system assets
- Reads, summarizes and sorts comments, applies the feedback to a new revision, and gives accessibility feedback or a critique from a chosen persona
- Parallel prompting on different frames, each with a chat that shows the steps taken, plus Stop and Undo
- Chats visible by default to teammates with a Full seat and edit access (chats can be made private), and the ability to follow collaborators' agents as they work on the canvas
- Custom skills written as Markdown files that follow the Agent Skills specification and run as slash commands, with skills shared in Figma Community
- Verified MCP connectors (including Notion, Linear, GitHub, Slack, Atlassian, Asana, Amplitude, Granola and Zapier) that read context and can write back, plus custom MCP connectors on paid plans
- Web search, file and image attachments, and search across other Figma files, FigJam stickies and Slides copy
- Builds shader effects and fills and generative plugins from a prompt, and adds motion and easing to designs
- Image tasks such as generating images, removing backgrounds, boosting resolution and converting static images to vector layers
Limitations
- Works only in Figma Design, in the browser or desktop app; Figma says it is not available on mobile or in other Figma products, and the FigJam and Slides agent is a separate closed beta
- Prototyping and interactions, vector and icon editing, and component slots are listed as coming soon, and the agent cannot export assets or create data-driven charts
- Since general availability on October 6, 2026, every prompt consumes AI credits, and usage varies with task complexity, attached context, connected libraries and chat history; Figma says it cannot show the cost before a prompt runs, and undoing a change does not refund its credits
- Credits are shared with Figma Make and other AI features, so heavy use of one can leave none for the agent until the monthly reset or until an admin buys more
- Seat credits reset monthly and do not roll over; extra credits can only be bought on paid plans, and subscription and pay-as-you-go prices are shown only in the admin dashboard, not on the public pricing page
- Only Full seats with edit access can have it edit shared team files; View, Dev and Collab seats can chat with it there but only edit in their own drafts
- Starter plan access is limited (150 credits a day, up to 500 a month), and library guideline files and custom MCP connectors require a paid plan
- Output is non-deterministic and needs designer review; the agent will not publish new assets to a shared library on its own
- No API for running the agent from outside Figma is documented; outside coding agents use the separate Figma MCP server instead
- Not available on Figma for Government, and on Enterprise plans admins can switch AI features off for the whole organization or for specific workspaces
Use cases
- Generating three distinct visual directions for a checkout or onboarding flow using the team's own component library
- Setting every chip or tile component across a file to a new state, or replacing lorem ipsum copy and placeholder images across a grid
- Writing component documentation that covers every state and variant, and auditing screens for hard-coded colors that should be tokens
- Turning a long comment thread or meeting notes from Granola into an organized list of changes and a revised version of the design
- Packaging a team's design-critique or accessibility checklist as a skill that anyone can run with a slash command
- Pulling a PRD from Notion or tickets from Linear or Jira into the chat to draft first-pass screens
Our take
The Figma agent's advantage is location. It works on the actual layers, libraries and comments your team already keeps in Figma, which makes it most valuable for jobs designers avoid: bulk component changes, consistent documentation, and turning scattered feedback into a revision. Teams that invest in a clean library, guideline files and a few shared skills will get the most out of it. The tradeoffs are cost and scope. Credit use per prompt varies and draws from the same pool as Figma Make, so pilot it with real tasks and check the AI credits dashboard before rolling it out widely. It also does not yet handle prototyping, vector work or export, so it speeds up the design file rather than replacing the rest of the workflow.
Who should use it
Product designers who want to explore several directions quickly with their own components, design-system teams that maintain and document large libraries, and product teams that want feedback from Figma comments, Notion, Linear or meeting notes applied directly in the file.
Who should skip it
Teams that do not use Figma, anyone who mainly needs working code or interactive prototypes, users on Starter or non-Full seats who expect to edit shared files heavily, and organizations that need a fixed cost per task or programmatic access to the agent.
Strengths
- Works directly on the layers of your Figma file instead of producing a separate image or code project
- Uses your published libraries, @-mentioned tokens and guideline files, so output follows your design system
- Strong at tedious bulk edits, design-system documentation and turning comments into changes
- Skills, MCP connectors and shared chats make workflows repeatable and visible to the team
- No separate subscription: it draws on the AI credits included with every seat, with admin controls for credit limits and AI access
Weaknesses
- Credit cost per prompt varies and is hard to predict; heavy users will need paid credit add-ons
- Prototyping, vector and icon editing, and asset export are not supported yet
- Only useful inside Figma Design; no standalone app or API
- Editing shared files requires a Full seat
Figma Agent pricing
Starter
Free
- Limited agent availability
- 150 AI credits/day, up to 500 credits/month
- Featured MCP connectors; no custom MCP connectors
Professional
$16/month per Full seat (billed annually)
Billed annual
- 3,000 AI credits/month per Full seat
- Dev seat $12/month and Collab seat $3/month (billed annually), each with 500 AI credits/month
- Monthly billing also offered, at a separate monthly price
- Custom MCP connectors and library guideline files
- AI credit add-on is a month-to-month subscription
Organization
$55/month per Full seat
Billed annual
- 3,500 AI credits/month per Full seat
- Dev seat $25/month and Collab seat $5/month, each with 500 AI credits/month
- Billed annually
- AI credit add-on is an annual subscription tied to the contract term
Enterprise
$90/month per Full seat
Billed annual
- 4,250 AI credits/month per Full seat
- Dev seat $35/month and Collab seat $5/month, each with 500 AI credits/month
- Billed annually
- Organization- and workspace-level AI on/off controls and credit billing groups
Free tier limits: On Starter, the agent has limited availability and AI credits are capped at 150 a day and 500 a month. View seats on paid plans have the same daily cap.
Note: The agent has no separate price; it consumes Figma AI credits, shared with Figma Make, Weave tools and other AI features. Credits per prompt vary with task complexity, context such as attached files, connected libraries, guidelines and connectors, and chat history; Figma's examples are about 30+ credits to change fonts, 100+ to summarize a canvas and 300+ to generate a five-screen mobile checkout flow. Seat credits reset monthly, do not roll over and cannot be shared. Admins can add plan-level credits through a subscription or pay-as-you-go billing with a monthly spending limit, and can set per-user limits on paid credits; add-ons are available only on paid plans, and current add-on prices are shown only in the admin dashboard. Professional can be billed monthly or annually; the $16 Full seat, $12 Dev seat and $3 Collab seat prices are the annual-billing rates shown on figma.com; monthly billing has its own price, which we could not confirm from Figma's site. The agent did not consume credits during the beta (May 20 to October 6, 2026).
Technical specs
Modalities
Text, Image, Design files
Where Figma Agent excels
Design system documentation at scale
Skills such as component anatomy and color-to-token mapping run in the file where the components live, producing documentation frames for every state and variant and flagging hard-coded values.
Applying crit and meeting feedback
The agent can sort comments by theme, pull notes from a connected tool such as Granola, annotate the canvas, and create a revised version of the design for review.
Fast exploration with real components
Parallel prompts on different frames generate several directions that already use the team's library, so designers compare and refine instead of rebuilding placeholders.
How Figma Agent fits alongside other tools
Figma Agent and Claude Code
Claude Code can write into Figma files through the Figma MCP server's write-to-canvas tool, which suits bringing designs from a codebase into Figma. The Figma agent runs inside Figma Design with extra context on your libraries and needs no MCP setup. Figma positions them as working together: code to canvas through the MCP server, then refinement with the agent.
Figma Agent and Lovable
Lovable turns prompts into full-stack web apps backed by editable code. The Figma agent produces and edits design layers in a Figma file, not running apps, so teams often design in Figma and build in a tool like Lovable or Figma Make.
Figma Agent and Krea Agent
Krea Agent produces images, video and audio assets by routing a brief across many generation models. The Figma agent works on structured UI designs with components and variables, and can pull in or generate images as one step of a layout.
Frequently asked questions
What is the Figma agent?
It is an AI agent built into Figma Design. You prompt it from the left rail or an on-canvas prompt box, and it creates and edits layers in the file: new screens and layouts, bulk changes to components, text and spacing, design system documentation, and revisions based on comments. It can use your connected libraries, skills and MCP connectors as context.
How much does the Figma agent cost?
There is no separate price. Since general availability on October 6, 2026, the agent uses Figma AI credits, which every seat includes each month: 3,000 for Professional Full seats, 3,500 for Organization, 4,250 for Enterprise, and 500 for Dev, Collab and View seats and Starter users (Starter and View are also capped at 150 a day). Credits per prompt vary; Figma's own examples are about 30+ for a font change, 100+ for a canvas summary and 300+ for a five-screen checkout flow. On paid plans, admins can buy more credits by subscription or pay-as-you-go.
Which Figma plans and seats can use the agent?
Figma's help center lists it on all plans, and the general availability announcement says it is available for Full seats on Professional, Organization, Enterprise and some Education plans, with limited availability on the free Starter plan (Figma's product page still says paid plans). In shared files, only Full seats with edit access can have it make edits; View, Dev and Collab seats can chat with it, and anyone can have it edit their own drafts. It is not available on Figma for Government, and Enterprise admins can turn AI features off for the organization or for specific workspaces.
What is the difference between the Figma agent, Figma Make and the Figma MCP server?
The agent edits design files in Figma Design. Figma Make generates interactive apps and prototypes as code. The Figma MCP server lets external agents such as Claude Code, Codex, GitHub Copilot or Cursor read Figma context and write into Figma files. Figma suggests using the MCP server to move work between code and the canvas, and the agent to refine it on the canvas.
Can the Figma agent build prototypes?
Not yet. Figma's help center lists prototyping and interactions as coming soon for the agent, along with vector and icon editing and component slots. For interactive prototypes from a prompt, Figma points to Figma Make.
Does the Figma agent train on my designs?
Figma says that if the AI content training setting is on, it may use content and usage data to improve Figma-specific models. Content training is on by default for Starter and Professional teams, which admins can switch off, and off for Organization and Enterprise plans.
Integrations & fit
About Figma Agent
Figma's agent lives in the left rail and in an on-canvas prompt box of Figma Design, in the browser and the desktop app. Instead of producing an image or a separate code project, it works on the layers of the file you are in: it creates frames and layouts, edits auto layout and spacing, swaps component variants, applies styles and variables, fills screens with realistic copy and images, renames layers, and reads and acts on comments. You can connect a published library and @-mention specific components, tokens and variables so its output uses your design system, and library owners can attach guideline files (Markdown, text or JSON) that tell the agent how the system should be used. Prompts can run in parallel on different frames, each with its own chat showing the steps it took, and chats are visible to teammates with edit access by default, so others can see the prompt and iteration history behind a design. Skills package a repeatable workflow as a Markdown file following the Agent Skills specification and run as a slash command, for example a component documentation or accessibility review routine. MCP connectors pull context from tools such as Notion, Linear, GitHub, Slack, Jira and Confluence, Asana and Granola into the chat and can write back when write tools are enabled, and paid plans can add custom MCP connectors. The agent can also search the web, attach files, build shader effects and generative plugins, and add motion. Figma launched it in beta on May 20, 2026 and made it generally available on October 6, 2026. Since then it draws from the same monthly AI credit allowance as Figma Make and other AI features, and Figma does not quote a fixed price per task: its own examples range from about 30 credits for a font change to 300 or more for a five-screen checkout flow. It is not Figma Make, which builds interactive code prototypes and apps, and not the Figma MCP server, which lets outside coding agents such as Claude Code or Cursor read and write Figma files. Figma's help center lists prototyping and interactions, vector and icon editing, and component slots as coming soon, and says the agent cannot export assets. It is most useful for teams with a maintained Figma library, and much less so for anyone who does not design in Figma.
Updates from Figma Agent
On October 6, 2026, Figma moved the agent in Figma Design (and Weave tools) from open beta to general availability. Prompts now draw from the same AI credit budget as Figma Make. The release also added library guideline files, search across Figma files, FigJam and Slides for context, and following collaborators' agents on the canvas.
Figma added the ability to ask the agent to author a skill in Figma, from a reference frame or a set of steps, preview it and use it in chat, and highlighted ten skills to start from.
At Config 2026 Figma made the agent available to more users in open beta and let it build generative plugins and shader effects and fills from a prompt. These live in a new Tools tab.
Figma introduced an agent built into the Figma Design canvas for exploring directions, bulk edits, design system work and acting on feedback. It rolled out gradually in beta and did not consume AI credits during the beta.
Are you the founder? Claim this listing →