Provides AI agents with structured access to real-world design patterns, UI blueprints, and semantic tokens for high-fidelity UI generation.
DesignIntelligence provides 5 well-named tools with explicit schemas and descriptions. All tools follow verb_noun naming (search_*, get_*), descriptions are substantial (100-200 chars), and input schemas include type information and parameter descriptions. However, output schemas are undocumented, the code shows return types in docstrings (e.g., 'list[dict]', 'dict') but no formal output schema specification. Error handling is minimal: search_design_patterns and get_design_blueprint can return error dicts but provide no recovery guidance. All tools are read-only, which is appropriate for this domain. The server loads external data (patterns.json, token_sets.json) and resolves indexed references transparently, which is sound design. Parameter defaults are sensible (platform='web', limit=5, detailed=False). Main gaps: (1) no documented output schemas, (2) no recovery hints in error responses, (3) no parameter value constraints beyond descriptions, (4) no batch variants despite patterns.json likely containing many items.
Get behavioral pattern for a design element or interaction
Get the full design blueprint for a specific pattern. Use after search_design_patterns to get deeper details on a specific example.
Get the full taxonomy of design categories available in the database. Use this to understand what page types, UX patterns, UI elements, industries, and visual styles you can search for.
Get semantic design tokens for consistent styling. Returns real hex color values, spacing scale, typography sizes, and border radii. Use these instead of hardcoding magic numbers.
Search the design pattern database for real-world UI examples. Use this BEFORE generating any UI to study how established products handle similar design challenges.
No documented output schemas. All tools have return types in docstrings (e.g., 'list[dict]', 'dict') but no formal field specification. LLMs cannot plan downstream tool calls or extract specific fields without guessing the response structure.
No enum constraints on filter parameters. 'page_type', 'industry', 'color_mode', and 'visual_style' accept free-form strings with only descriptive text examples (e.g., 'e.g. Dashboard, Pricing, Onboarding'). LLMs will hallucinate invalid values.
Error responses lack recovery guidance. get_design_blueprint returns {'error': '...'} on not found but does not suggest calling search_design_patterns() or provide available pattern IDs. Agents cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 17 | - | v1 |
get_behavioral_pattern description is minimal (55 chars, below 100-char baseline) and vague. Does not explain what a 'behavioral pattern' is, when to call it instead of get_design_blueprint, or how to discover valid pattern names. Parameter 'pattern_name' has no guidance on format or valid values.
No pagination support. search_design_patterns accepts a 'limit' parameter but no offset/page/cursor parameter. If users want results beyond the first 5, there is no mechanism to retrieve page 2. Violates pattern:paginated-result.
No batch variants. If an agent needs to fetch multiple design blueprints or semantic tokens in a loop, it must call get_design_blueprint or get_semantic_tokens repeatedly. No get_design_blueprints (plural) tool to fetch multiple patterns in one call.