Better Way of Visual Intelligence — Agent-native design decision protocol. MCP server for design analysis, generation, critique, and learning.
BWVI exposes 15 tools with complete JSON schemas and descriptions, but quality is uneven. Naming follows verb_noun convention (analyze_design, generate_design, critique_design) which is good. However, descriptions are often terse (10-50 chars) and lack actionable context for LLM selection. Parameter descriptions exist but are minimal. Critical issue: generate_design marks 'direction' as deprecated in the description but still includes it in the schema, creating ambiguity. Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent; no recovery guidance or categorization. The learn_system_* tools (status, diagnose, ingest, knowledge) form a subsystem but lack clear composition guidance. Overall, definitions are present but fall short of production quality.
Analyze design task → direction recommendations
Inject CSS animations into HTML
Get full brand details (colors + typography)
Critique HTML → 10-dimension quality score
Compare two HTML files → metric diff report
Generate HTML design from task + confirmed decisions
Extract design tokens from URL
Output schemas not documented. LLMs cannot predict response structure for any tool. Responses are unstructured or inferred from implementation.
Deprecated parameter 'direction' in generate_design still present in schema despite description saying '已废弃'. Creates ambiguity: should LLM use direction or confirmed_decisions.direction? Violates single-source-of-truth principle.
No error handling or recovery guidance. Tools provide no categorization (retryable vs user-fixable vs fatal) or actionable error messages. LLMs cannot self-correct on failure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Diagnose weakest capability bottleneck
Ingest knowledge from URL into learning system
List all learned knowledge chunks
Get learning system capability status report
Find matching blueprint for a task
Search/list 115 built-in brand systems
List all 10 design directions
List all 56 visual styles
Descriptions are terse (10-50 chars) and lack context for LLM selection. E.g., 'Analyze design task → direction recommendations' does not explain WHEN to call this vs other tools or what 'direction recommendations' means.
learn_system_* subsystem (status, diagnose, ingest, knowledge) lacks composition guidance. No documentation of call order, dependencies, or how results chain together. LLMs must infer workflow.