Server has 6 tools with basic schemas and Japanese descriptions. Critical gaps: (1) EmptyInput tools (get_drawing_info, list_layers) lack parameter descriptions entirely, they accept no params but provide no guidance on what they return. (2) Descriptions are in Japanese, limiting LLM comprehension in English-first environments. (3) No output schemas documented, LLMs cannot predict response structure. (4) run_script tool is DESTRUCTIVE but lacks confirmation/dry-run pattern. (5) Parameter descriptions exist for shape tools but are minimal (e.g., 'Start point X coordinate' lacks units, range, or coordinate system context). (6) No error handling guidance, tools return raw strings from bridge, no structured error classification. (7) Tool names are clear (verb_noun pattern) but descriptions do not explain WHEN to use each tool or dependencies between them.

• Add English descriptions (or bilingual) for all tools. Current Japanese descriptions are inaccessible to English-speaking LLMs. Aim for 50 - 200 characters per tool, explaining WHAT it does, WHEN to use it, and any prerequisites.
• Document output schemas for all tools. Specify what fields get_drawing_info and list_layers return (e.g., {drawing_name: string, width: number, height: number, layers: [{name: string, visible: boolean}]}). This enables LLMs to chain tools and extract relevant data.
• Add parameter descriptions with units and ranges. E.g., 'x1: Start point X coordinate in millimeters (0 - 10000)' for create_line. Include coordinate system context (e.g., 'origin at bottom-left').
• Implement confirmation/dry-run for run_script. Add a 'dry_run' boolean parameter (default false) or a separate 'preview_script' tool. Require explicit confirmation before executing arbitrary code.
• Replace unstructured error strings with structured error objects: {error_code: string, message: string, recovery_hint: string, retryable: boolean}. E.g., 'Connection error: Vectorworks plugin not loaded. Recovery: Load the VWMCPBridge plugin in Vectorworks and retry.' This guides LLM recovery.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry