Loading ToolBench
MCP server that opens claude.ai/design for UI/UX design sessions with context passed via clipboard
Both tools have adequate descriptions (60-90 chars, within baseline 34-392 range) and clear naming with action verbs. However, critical gaps undermine quality: (1) Input schemas lack descriptions for parameters, 'description' and 'project_type' in open_design_session have no parameter-level docs; 'component' and 'style_notes' in design_component similarly lack parameter descriptions. (2) No output schema documented, tools return plain strings, but LLMs need to know what fields/structure to expect for downstream chaining. (3) No error handling guidance, if the 'open' command fails or pbcopy is unavailable (common on non-macOS), the tool silently fails with no recovery path. (4) No input validation, 'project_type' accepts any string, not constrained to [web, mobile, desktop] as the description claims. (5) No idempotency or side-effect clarity, tools open URLs and modify clipboard, but descriptions don't warn that repeated calls have side effects. These are foundational gaps that prevent confident LLM use.
Open claude.ai/design to design a specific UI component such as a hero section, nav bar, pricing table, or footer.
Open claude.ai/design for a new UI/UX design session. Call this when starting work on any website, app, or interface that needs visual design help.
Input parameters lack descriptions. 'description', 'project_type', 'component', and 'style_notes' have no parameter-level documentation. LLMs cannot infer what values are valid or what each parameter controls.
No output schema documented. Tools return plain strings ('Opened claude.ai/design...'), but LLMs need structured output definitions to plan downstream actions and extract data reliably.
No input validation or enum constraints. 'project_type' description claims 'web, mobile, or desktop' but accepts any string. No validation in code. LLMs will pass invalid values without feedback.
No error handling or recovery guidance. If 'pbcopy' or 'open' commands fail (e.g., on Linux, Windows, or headless systems), the tool silently fails with no actionable error message for the LLM.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Side effects not documented. Tools open URLs and modify clipboard state, but descriptions do not warn that repeated calls have non-idempotent side effects (e.g., clipboard overwrite).