Refreshing decks, boards, and prototypes through the Figma MCP server
Two months after opening the canvas to AI agents, the Figma MCP server has expanded across more of the platform than before. It now covers Figma Slides, FigJam, Figma Make, and the new Figma agent—so presentation decks, FigJam boards, and Make prototypes can be created or updated from a prompt. The server also supports custom fonts and, through the new download_assets tool, lets you pull images and icons from design files as SVG, PDF, JPG, or PNG.
Here are four workflows Figmates are running today with those capabilities.
Deck maintenance without the busywork
Mallory Dean, a designer advocate at Figma, keeps an evergreen deck about Figma’s AI product launches that she refreshes every few weeks. Because the MCP server now supports uploads of custom fonts—any typeface saved locally can render correctly—her prompt in the code editor could go straight at the content rather than the formatting.
After the Figma agent launched, she prompted: "Update my Figma AI deck to include our new Figma agent. Pull content from Slack, Google Drive, our Shortcut blog, and Release Notes webpage. Give me suggestions for what to refresh, plus ideas for new slides." The agent gathered the relevant conversations and launch material, then used the use_figma tool and the /figma-use-slides skill in Figma Slides to work against her template.
The slides still needed a human pass for images, placeholders, and copy edits, but most of the content work was done before she started. With custom font support, the agent rendered type from her uploaded fonts rather than web-safe substitutes, preserving the deck’s branding.
The same setup works across roles: a PM assembling a kickoff deck, a designer showing an exploration, marketing writing a GTM plan, or sales refreshing a customer deck. Output comes from your team’s design system, so it stays on-brand by construction.
From live project data to a ready FigJam board
Skills let agents produce more consistent results. You can browse custom skills in the Figma community or submit your own to the Figma community-resources repo for others to use with the MCP server.
Prasant Lokinendi, a product manager at Figma, runs feature kickoff workshops often. Preparing an engaging FigJam from scratch—collecting company context and structuring sections for the session—takes time, so he wrote /figjam-builder, a custom skill that encodes those instructions and removes the need to re-prompt each time.
For the Make voice-to-text launch earlier this year, he prompted his agent to generate a FigJam board using project context from Slack, Asana, and Notion, plus analytics from Hex. Rather than starting from a blank board, he worked from current information including product vision, customer insights, and key decisions.
Closing the loop between canvas and production code
MCP support in Figma Make lets design edits flow toward a production pull request without leaving Figma. Iris Lin, a product designer at Figma, uses that loop on real product work.
When Iris and a teammate built a sample audio editor demo, her teammate shipped an initial version, but Iris wanted to refine the designs. Interactions are hard to demonstrate in a static file, so she branched the code and built the working version in Figma Make: draggable audio clips, a popover with level controls, and a scrubbing playhead.
Wanting to edit her design system for the demo, she prompted in the Make prompt box: “Can you bring back the preview here into Figma as design layers?” The screen came onto the canvas reconstructed from her library’s components. Iris then updated the audio clip component following her normal canvas patterns, adding distinct default, hover, and drag states. She sent it back the other way with “Pull those new states back into the code.” The agent read her changes and wrote all three states into the component, ready to push to GitHub as a PR.
The workflow keeps control over what ships while staying inside Figma. The agent reads and writes your real components on both ends, similar to how Code Connect maps libraries to production code—so design decisions that once lost fidelity between handoff and review now carry through to the pull request.
Pushing a code-only screen into the design canvas
Yarden Katz, the product manager behind the Figma MCP server, regularly tests a workflow that starts with a screen that exists only in code. The objective: bring it into Figma, attach it to the right design system, and hand it to designers.
She works from a sample app containing a login flow and dashboard, using the MCP server to pull assets directly from Figma via the new download_assets capability. Unlike a screenshot, this returns the actual exportable file — SVG, PDF, JPG, or the original source image — with no manual export required.
From her code editor, she prompts: "Push this dashboard and the login flow into Figma. Reuse my existing components and variables where they exist, and generate proper component sets and variables where they don't." The Figma plugin ships with built-in skills that give the agent context on how Figma works. It reads the library in both Figma and the codebase and decides what to reuse versus what to create from scratch, rather than duplicating completed work.
Where the agent takes over
The result is a strong first pass, not a finished design. Auto layout, fonts, and unmapped colors still require attention. At that point the Figma agent steps in, working directly on the canvas with access to the design system's context. Katz prompts it to fix the layout, correct typography, and map every color to the proper variable. Once she is satisfied with the result, the work is pushed back to code through the MCP server. The download_assets tool remains available for pulling source images and icons directly from the agent, eliminating separate export steps.
This workflow is essentially a tag-team arrangement. The MCP server connects Figma to the agent already running in the developer's environment and translates a code-only screen into a canvas built from real components. The Figma agent then refines the result natively, exploring variations while staying grounded in the design system's components and tokens.
Figma Make's production codebase integration and the Figma agent are currently in closed beta. Write capabilities in the Figma MCP server are available in open beta, with a setup guide for getting started.
