Figma Opens Canvas to AI Agents with MCP Server Beta

Figma has launched a beta for its MCP server, enabling AI agents to read and write directly to Figma files, leveraging established design systems, components, and variables. This bridges the gap between code and canvas, allowing seamless movement in workflows—agents can generate designs from codebases, edit layers, and ensure outputs align with team standards.

Key features and tools include:

•  use_figma  tool for agents to operate on the canvas using design systems.

•  generate_figma_design  for converting HTML/UI into editable Figma layers.

• Skills as markdown instructions that guide agents on workflows, conventions, and context—e.g.,  /figma-generate-library  to create components from code,  /apply-design-system  to connect designs to systems.

Nine community-contributed skills launch today, such as  /create-voice  for screen reader specs (from Uber),  /cc-figma-component  for JSON-to-components (One North), and  /multi-agent  for parallel workflows (Augment Code). Agents like Claude Code, Codex, Cursor, and Warp already support this, with self-healing loops for iterative refinements via screenshots and auto-layout interactions.

The feature is free during beta (future usage-based pricing) and enhances surfaces like Code Connect, Figma Draw, and FigJam via Plugin API parity goals.

Why This Matters:

This development democratizes advanced design workflows by empowering AI agents with real design context, reducing generic outputs and accelerating product iteration. It fosters collaboration between designers and developers, embedding team knowledge into AI processes for higher-quality, consistent products. Ultimately, it positions Figma as a central hub for agentic design, unlocking efficiency gains for teams at companies like OpenAI and Uber.

Source: Figma's editorial newsletter