Loading ToolBench
Bridge Figma to any platform — Unity UGUI mappings, design tokens for React/Flutter/SwiftUI/Tailwind, accessibility audits, and AI-optimized specs through MCP
Server has 11 well-named tools with complete input schemas and structured output types. Tool names follow verb_noun convention (inspect_layout, extract_design_tokens, map_to_unity). All tools have descriptions (194 - 250 chars, within baseline 34 - 392 range). Input parameters are typed and mostly described. However, descriptions lack actionable context for LLM selection (no 'when to use' guidance, no dependency hints). Output schemas are documented in TypeScript interfaces but not exposed in tool registration metadata. Error handling is minimal, no recovery guidance, no categorization of retryable vs fatal errors. Security concern: access_token exposed as a required parameter in all 11 tools, violating secret-injection pattern. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only.
Audits a Figma frame for WCAG 2.1 accessibility issues: contrast ratios, touch targets, missing alt text, font sizes. Returns issues with severity levels and an overall accessibility score.
Scans Figma components and a local codebase, then maps component names to likely implementation files using exact, case-insensitive, and partial filename matching.
Fetches two Figma file versions and reports added, removed, and modified nodes by comparing names, types, geometry, fills, and style properties.
Exports Figma nodes as images (PNG, JPG, SVG, PDF) via the Figma image export API. Returns download URLs for each requested node.
Extracts design tokens (colors, typography, spacing) from a Figma file and exports them in multiple formats: style-dictionary, css-variables, or tailwind configuration.
access_token exposed as required parameter in all 11 tools. Credentials must never appear in tool parameters, use server-side secret injection via environment variables or vault. Agent traces log every parameter, leaking secrets into logs and prompt history.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all 11 tools being read-only. Tool annotations enable agents to reason about safety and retry behavior. All tools should declare readOnlyHint: true.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Extracts prototype flows from a page or frame by finding transition links in the node tree, then returns directed frame-to-frame connections and a deterministic traversal order.
Extracts variant information from a Figma component set, including variant properties, dimensions, layout configuration, and styling for each variant.
Inspects a Figma frame's layout structure, including auto-layout configurations, constraints, text runs, and accessibility warnings. Returns a hierarchical node summary with positioning modes and framework hints.
Maps a Figma frame to Unity UGUI components, generating RectTransform configurations, LayoutGroup settings, and component suggestions with confidence levels.
Resolves component instances in a Figma file to their source components, returning component names, keys, and descriptions.
Simplifies a Figma node's context by extracting essential layout, styling, and text information, optimized for token-constrained AI contexts. Supports framework-specific simplification (web, react, unity, swiftui).
Descriptions lack actionable context for LLM selection. No 'when to use' guidance, no dependency hints, no distinction from similar tools. E.g., extract_design_tokens, extract_variants, and extract_flows all start with 'Extracts', LLM cannot easily distinguish when to call each.
No error handling guidance. Tools do not document what errors are retryable, what requires user intervention, or how to recover. E.g., 'file not found' vs 'rate limited' vs 'invalid token' all need different agent responses.
Output schemas documented in TypeScript interfaces but not exposed in tool registration. LLMs cannot see what fields to expect from responses, forcing them to guess or make extra discovery calls.