Server has 15 tools with mixed quality. Most tools have descriptions and basic schemas, but critical gaps exist: (1) Many parameter descriptions are in Japanese, creating LLM comprehension barriers. (2) Several tools lack complete input schemas or have minimal descriptions. (3) Tool naming is inconsistent, some follow verb_noun pattern (generate_pdf_sync, download_file) while others are vague (fetch, search). (4) No documented output schemas for any tool. (5) Error handling guidance is absent. (6) authenticate tool exposes OAuth flow details in description but lacks security context. Average tool score: 52/100.

• Translate all parameter descriptions to English. Replace Japanese descriptions with clear English equivalents: 'designId: Template design ID in UUID format' instead of 'デザインID（UUID形式）'.
• Document output schemas for all tools. Add a 'Returns' section to each tool description specifying field names, types, and meanings. E.g., 'Returns: {templates: [{id: string, name: string, category: string}], total: number}'.
• Rename 'search' to 'search_templates' and 'fetch' to 'fetch_api_data' or remove 'fetch' entirely if it's a generic HTTP wrapper. Domain-specific names help LLMs select the right tool.
• Add error recovery guidance to tool descriptions. E.g., 'If designId is invalid, call list_templates() to discover valid IDs.' or 'On auth failure, call authenticate() with force=true.'
• Standardize parameter naming: use 'template_id', 'design_id', 'gallery_slug' consistently. Add type suffixes (_id, _slug, _name) to disambiguate. Document which parameters accept human-readable names vs system IDs.

Inferred effective spec: 2026-07-28+.

• Relies on Roots (deprecated) - pass files/dirs via tool parameters, resource URIs, or configuration