Single tool 'search_word' has a basic schema with Zod types and descriptions, but critical gaps undermine production readiness. Tool description is minimal (12 chars: '단어를 검색합니다.'), far below the 50-200 char LLM-optimized baseline. Parameters lack actionable constraints (no min/max for numeric start/num, no enum for query format). Output schema is undocumented, LLMs cannot predict response structure. Error handling returns raw JSON without recovery guidance. No pagination limits stated despite API returning potentially large result sets. Tool does not follow verb_noun naming (search_word is acceptable but description is too terse to guide LLM selection). Overall, this is a minimal wrapper around a Korean dictionary API with insufficient metadata for reliable agent use.

• Expand tool description to 80-150 characters explaining: (1) what it does ('Searches the National Institute of Korean Language Standard Dictionary'), (2) when to use it ('Use this to find definitions, parts of speech, and usage examples for Korean words'), (3) what it returns ('Returns matching words with definitions and grammatical info'). Reference pattern:tool-description.
• Document the output schema explicitly in the tool description or as a separate schema field. Describe that response contains: channel.total (count), channel.item[] (array of words), each item has word, pos (part of speech), sense[] (definitions). This enables LLMs to extract and chain data.
• Add constraints to numeric parameters: start (min: 0, default: 0), num (min: 1, max: 50, default: 10). Cap results at 50 items and document this limit in the description to prevent context window exhaustion.
• Mark 'query' as required (remove .optional()) since the tool throws an error if omitted. Update schema to reflect actual behavior.
• Enhance error messages with recovery guidance. E.g., 'Search term required. Provide a Korean word or phrase.' or 'API returned no results for "[query]". Try a shorter term or check spelling.' Reference pattern:recovery-guide.