Loading ToolBench
Workshop resources and tools for Design-to-Code AI training, including Figma-to-code automation, Azure AI integration, and component generation utilities
Scoring was not performed
9 of 10 tools have no visible input schema. Tool definitions are inferred from npm scripts in package.json with no formal MCP schema registration visible in source code.
Tool naming violates verb_noun convention. Uses colon-prefixed names (test:mcp-connection, generate:batch) instead of action verbs (test_mcp_connection, generate_batch). LLMs struggle to infer intent from non-standard naming.
Descriptions for 7 tools are under 20 characters (generate:batch='Batch generates...', extract:tokens='Extracts design tokens...') or trivially short. Insufficient for LLM tool selection guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 23 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 25 | - | v1 |
No output schemas documented for any tool. LLMs cannot plan downstream calls or understand what data they'll receive.
No error handling guidance visible. No documentation of what errors each tool can return, whether they're retryable, or how the LLM should recover.
For tools that write state (generate:component, generate:batch, generate:from-image), descriptions do not explicitly warn about irreversible side effects or offer dry-run capability.
Parameter relationships undocumented. For figma.getComponent, includeStyles and includeVariants have no documentation of dependencies or interactions. Missing guidance on what happens if both false.
No evidence of input validation or constraint declaration. Parameters like targetDir (in generate:component) have no documented format, length limits, or valid path patterns.