Server has 5 tools with basic descriptions in Japanese, but significant gaps in schema completeness, parameter documentation, and error guidance. Tool names follow verb_noun pattern (upload_*, replace_*, post_*, update_*, get_*), which is good. However, input schemas lack proper type definitions for several parameters (e.g., 'labels' array items not typed, 'fields' array items not typed). Descriptions are present but often generic and lack actionable guidance for LLM selection. Error handling is verbose but does not classify errors as retryable/user-fixable/fatal. No output schemas documented. Parameters like 'target_blog' accept arbitrary strings instead of enums, inviting hallucinated values.

• Convert 'target_blog' parameter to enum: {"type": "string", "enum": ["default", "site1", "site2"], "description": "Target blog identifier..."}. Prevents hallucinated values.
• Add item type to array parameters: 'labels' → {"type": "array", "items": {"type": "string"}, "description": "Array of label strings..."} and 'fields' → {"type": "array", "items": {"type": "string", "enum": ["id", "title", "status", "url", "labels", "content", "published", "updated", "replies", "author"]}}.
• Document output schemas for each tool. E.g., upload_image_to_gcs returns {"type": "object", "properties": {"url": {"type": "string"}, "status": {"type": "string"}}}. Enables LLM to chain calls.
• Add error classification to post_to_blogger and update_blogger_post. Prefix error messages with [RETRYABLE], [USER_FIXABLE], or [FATAL] so LLM knows next action.
• In parameter descriptions for post_to_blogger and update_blogger_post, explicitly state: 'Mutually exclusive: provide either content OR file_path, not both.' Prevents ambiguous calls.