Server has 6 tools with explicit schemas and descriptions, but quality is inconsistent. All tools are READ_ONLY and well-intentioned, but descriptions are in Korean (non-standard for LLM-facing tools), parameter descriptions lack detail on constraints/formats, and no output schemas are documented. Tool names follow verb_noun pattern (list_, get_, search_) which is good. However, descriptions average ~80 chars (below 194-char baseline) and lack actionable guidance on when to use each tool vs. alternatives. No error handling guidance, no pagination support despite API returning lists, and no documented output structure for LLMs to plan downstream calls.

• Translate all tool descriptions and parameter descriptions to English. Use clear, concise language optimized for LLM comprehension (50-200 chars per description). Example: 'list_datasets' → 'List all available AI Hub datasets with their IDs and titles. Use this to discover datasets before calling get_dataset_details.'
• Document output schemas for each tool. Define what fields are returned (e.g., list_datasets returns array of {dataSetSn, dataSetNm, fldNm}). This enables LLMs to extract dataset_key for chaining into get_dataset_details.
• Add pagination to list_datasets and list_data_packages. Accept 'limit' (default 20, max 100) and 'offset' parameters. Return total_count and next_offset so LLMs can iterate through large result sets without context explosion.
• Enhance parameter descriptions with format/constraint details. Example: 'dataset_key: String. The numeric AI Hub dataset identifier (e.g., '593'). Obtain valid keys by calling list_datasets first.' Add enum or pattern constraints where applicable.
• Implement structured error responses. Return JSON with {error_code, message, recovery_hint}. Example: {error_code: 'DATASET_NOT_FOUND', message: 'Dataset 999 does not exist', recovery_hint: 'Call list_datasets to see available datasets'}.