Built this because the official Figma MCP get_design_context was too long and noisy to pass directly into coding agents. Instead of forwarding huge blobs of metadata, coordinates, repeated classes, and SVG output, this bridge sits in the middle and converts selected Figma nodes into a smaller handoff package: compact Markdown screenshot path instead of base64 preserved design tokens deduped component structure diff handoff for previously seen components local component registry hints Ollama-based pre-analysis Current benchmark on the DashStack Dashboard fixture: official raw: 13,926 input tokens bridge handoff: 6,733 input tokens reduction: 51.65% Goal wasn’t just reducing tokens. It was making UI implementation outputs more stable by passing only the information the coding agent actually needs. Works with React, Vue, Svelte, and HTML targets, plus multiple styling profiles. Repo and demo are in the launch page. Curious how others are handling Figma → LLM UI workflows.

figma-cost-optimizer-bridge

Turn noisy output into compact handoff docs for coding agent

Turn noisy output into compact handoff docs for coding agent