An MCP server that treats Figma like a database

What it does

Figma Console MCP is a Model Context Protocol server that sits between AI clients and Figma. It exposes over a hundred tools for reading design data, manipulating frames and components, managing variables, and exporting tokens to DTCG JSON or CSS. The server can run locally via a Desktop Bridge plugin, in a cloud-relay mode for web AI clients, or as a read-only remote endpoint.

The interesting bit

The project treats design artifacts as versioned infrastructure: it can diff Figma file snapshots, generate markdown changelogs, and binary-search through history to find when a variant first appeared. It also pushes back—letting you edit code-side tokens and sync the changes upstream into Figma, which replaces the usual one-way export pipeline.

Key highlights

• 106 tools in full local mode, including write access for frames, components, and design tokens
• Bidirectional token sync: exports Figma variables to DTCG JSON/CSS and imports code-side edits back into Figma
• Accessibility scanning with 14 WCAG design checks and axe-core code scanning
• Version history with binary-search blame to trace when properties or variants were introduced
• Four connection tiers: full local (NPX/Git), cloud relay (95 tools, no Node.js), and read-only SSE (9 tools)

Caveats

• Full write access and real-time monitoring require the Desktop Bridge plugin and Figma Desktop; the read-only remote mode is genuinely read-only
• The README notes that Figma caches plugin files aggressively, so re-importing the bridge plugin is occasionally required after updates
• Claude Code’s native SSE transport has a known bug, requiring a workaround with mcp-remote

Verdict

Worth a look if you want AI agents to operate directly on your design system instead of just reading screenshots. Skip it if you only need occasional Figma data extraction—nine read-only tools is a thin offering for that use case.