@siu-issiki/fig-mcp
v2.0.0
Published
MCP server for parsing .fig files
Maintainers
Readme
fig-mcp (siu-issiki fork)
MCP server for parsing .fig files. Enables AI assistants to understand and extract design information from the .fig file format for implementation guidance.
Forked from bilalba/fig-mcp with substantial rendering-fidelity and robustness improvements — see Fork changes.
Installation
No install step needed — npx fetches and builds directly from GitHub
(the first run takes a minute; afterwards it runs from the npx cache):
npx -y github:siu-issiki/fig-mcp --helpFor development:
git clone https://github.com/siu-issiki/fig-mcp
cd fig-mcp && npm install && npm run buildQuick Start
Add to Claude
claude mcp add fig-mcp -- npx -y github:siu-issiki/fig-mcpOr with a local clone:
claude mcp add fig-mcp -- node /path/to/fig-mcp/dist/index.jsThen ask Claude to parse your .fig files:
"Parse my design.fig file and show me the document structure"
Web Viewer
Browse and preview .fig files in your browser:
fig-mcp viewer design.fig
# Opens http://localhost:3000Features:
- Tree navigation with collapsible nodes
- SVG preview with zoom/pan
- Node details panel
- Copy node IDs for MCP tool calls
CLI Inspector
Inspect .fig files from the command line:
fig-mcp inspect design.fig summary # Show document structure
fig-mcp inspect design.fig stats # Show node type counts
fig-mcp inspect design.fig list # List archive contents
fig-mcp inspect design.fig json # Output simplified JSONCLI Commands
| Command | Description |
|---------|-------------|
| fig-mcp | Start MCP server (for AI assistants) |
| fig-mcp viewer <file> [port] | Open web viewer |
| fig-mcp inspect <file> [cmd] | Inspect file |
| fig-mcp --help | Show help |
| fig-mcp --version | Show version |
MCP Tools
The MCP server exposes the following tools for AI assistants:
Document Structure
| Tool | Description |
|------|-------------|
| parse_fig_file | Parse and return simplified document structure |
| get_document_summary | Text tree of document structure with pagination |
| get_tree_summary | Hierarchical summary for drill-down navigation |
| list_pages | List all pages (canvases) in the document |
| get_page_contents | Get contents of a specific page |
Node Queries
| Tool | Description |
|------|-------------|
| find_nodes | Find nodes by type or name |
| get_node_details | Get details for a node by path |
| get_node_by_id | Get details for a node by GUID |
| get_layout_info | Get inferred layout properties |
Content Extraction
| Tool | Description |
|------|-------------|
| get_text_content | Extract all text content |
| get_colors | Extract unique color palette |
| list_nodes_with_fills | List nodes with fill paints |
Image & Rendering
| Tool | Description |
|------|-------------|
| list_images | List all images with metadata |
| get_image | Get image by hash (base64) |
| get_thumbnail | Get document thumbnail |
| render_screen | Render node subtree as PNG (see options below) |
| get_vector | Export vector as SVG, PDF, PNG, or WebP |
Debugging
| Tool | Description |
|------|-------------|
| get_schema_info | Kiwi schema information |
| get_raw_message | Raw decoded message |
| list_archive_contents | List files in the archive |
| clear_cache | Clear file cache |
render_screen options
| Option | Description |
|--------|-------------|
| includeImages | Embed image fills (default: false) |
| downloadFonts | Download missing Google Fonts for fallback text, cached in ~/.cache/fig-mcp/fonts (default: true; sends font family names to Google — set false for fully offline rendering) |
| fontMap | Fallback families for non-Google fonts, e.g. {"AFSGillSBCond": "Gill Sans"} |
| fontDirs | Extra directories to scan for font files |
| scale, maxWidth, maxHeight, background, maxDepth, includeText/Fills/Strokes/Shadows | Rendering controls |
Most text renders from glyph outlines embedded in the file and needs no fonts at all; the font options only affect text without embedded glyph data.
How It Works
.figfiles are ZIP archives containing:canvas.fig- Main document data (kiwi binary format)meta.json- File metadatathumbnail.png- Preview imageimages/- Image assets
The
canvas.figuses Evan Wallace's kiwi binary formatThe kiwi schema is embedded in each file and extracted at parse time
Document data is decoded and transformed into structured information
Layout properties are inferred from node positions and auto-layout settings
Features
- Parse
.figfiles locally without API access - Extract document structure, nodes, and hierarchy
- Infer layout properties (flexbox-like direction, gap, padding, alignment)
- Extract colors, text content, and styling information
- Render nodes to PNG screenshots with near-design fidelity
- Export vectors as SVG, PDF, PNG, or WebP
- Effects (shadows, layer blur), gradients, masks, dashed borders — background/glass blur is approximated by the translucent fill (SVG has no backdrop-filter)
Fork changes
Rendering fidelity (verified pixel-by-pixel against Figma prototype screenshots):
- Embedded glyph rendering: text renders from the glyph outlines stored in the file (
derivedTextData.glyphs), reproducing exact letterforms of fonts that are not installed anywhere — including circular text on paths with per-glyph rotation - Component INSTANCE resolution in both rendering and
get_text_content(SYMBOL expansion with overrides, incl.componentPropAssignments) - Masks become SVG clipPaths (icon "Bounding box" layers no longer paint over glyphs)
- Frame borders render from Figma's precomputed
strokeGeometry(dashes included; invisible borders stay invisible) - Linear/radial gradient fills, mirrored (negative-scale) nodes,
textCase, rotated text - Sibling z-order follows the fractional-index position (not nodeChanges order)
- Google Fonts on-demand download +
fontMap/fontDirsfor fallback text
Robustness:
- Schema-driven argument validation with clear error messages
resolveNodePathhandles node names containing/and reports candidates on failure- BigInt-safe
get_raw_message; graceful fallbacks when geometry blobs are missing server.tssplit into per-category tool modules; vitest suite (npm test)
Testing
npm testUnit tests run standalone. Integration tests against a real file run when
FIG_TEST_FILE points at a .fig export (defaults to ~/Downloads/toritori2.0.fig).
Requirements
- Node.js 20 or higher
Limitations
- The
.figformat is undocumented and may change - This is for local
.figfiles only (use a cloud API for hosted files) - Some complex properties may not be fully parsed
License
MIT
