Four tools with complete schemas and descriptions. Naming follows verb_noun convention (search_, get_, refresh_). Descriptions are adequate (50-100 chars) but lack actionable context for LLM selection. Parameter schemas are well-typed with enums and constraints. Output structure is JSON but not formally documented. Error handling exists (isError flag) but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Missing dependency hints and per-parameter descriptions for some fields.

• Add tool annotations: registerTool() calls should include readOnlyHint: true for search_entries, get_entry, dictionary_status; destructiveHint: true for refresh_dictionary. This enables LLM reasoning about retry safety and side effects.
• Document output schemas explicitly. For search_entries, declare that response.content[0].text is JSON array of {target_code, word, pos, definition, ...}. For get_entry, declare single entry object structure. This enables LLMs to extract and chain results.
• Enhance refresh_dictionary description: 'Downloads the official Korean dictionary dump and rebuilds the local SQLite database. This operation is destructive (overwrites existing data), may take 5-10 minutes, and is not safe to call concurrently. Returns status object with import_count and timestamp.'
• Add recovery guidance to get_entry error: 'Entry {target_code} not found. Try search_entries(query=partial_word) to find similar entries, or call dictionary_status() to verify the database is initialized.'
• Improve fields parameter descriptions with use-case hints: 'Array of field names to include. Common choices: ["word", "pos", "definition"] for quick lookup; ["word", "definitions", "examples", "etymology"] for detailed study. Omit for faster results.'
• Add dependency hints to search_entries: 'If you only have a target_code, use get_entry() directly. If you have a partial word, use this tool with match="prefix" or match="contains".'
• Document pagination behavior: search_entries accepts limit (1-100, default 20) and offset (default 0). Clarify whether total_count is returned to enable LLM-driven pagination loops.