Two tools present with explicit schema registration in src/index.ts. Tool names follow verb_noun convention (save_word, get_words_list). Descriptions exist but are generic and lack crucial implementation details (e.g., what constitutes 'success', what the API returns beyond a textual response, error conditions). Input schemas are properly typed JSON Schema with some parameter descriptions, but output schemas are NOT documented, LLMs cannot predict response structure. Error handling in the code returns generic strings without actionable guidance. Parameter constraints present (star 1-5, page_size 1-100, pagination), which is good. However, the descriptions average ~80 chars and lack LLM-optimized guidance on when to use each tool or what data dependencies exist. No mention of dependencies (e.g., 'requires Eudic auth token set in env'). Security: Auth token via environment variable is correct, but no documentation in tool descriptions.

• Expand tool descriptions to 80-150 characters, answering: WHAT the tool does (beyond the name), WHEN to call it (vs. alternatives), and WHAT it returns (structure, not just count). Example: 'save_word saves an English word with optional star rating (1-5) and category tags to your Eudic study list. Returns {id, created_at, word}. Use this after encountering a new word; use update_word if modifying an existing entry.'
• Document explicit output schemas for both tools. E.g., save_word returns {success: boolean, word_id: string, saved_at: ISO8601, errors?: string[]}. get_words_list returns {words: [{word: string, star: number, added_at: ISO8601}], page: number, total: number, has_next: boolean}. Include these in tool metadata or a separate schema document.
• Enhance error messages with actionable recovery guidance. Instead of 'Error saving word', return: 'Failed to save word: Invalid authentication token. Set EUDIC_AUTH_TOKEN env var and restart the server.' or 'Word already exists in category 5. Call update_word(word_id=X, star=4) to modify.'
• Document parameter relationships. In save_word, clarify: 'category_ids and star are independent; pass both to tag the word with multiple categories AND rate its difficulty. category_id=0 means uncategorized.'
• Add dependency hints in descriptions. E.g., 'If you only have the word text (not its ID), call this tool. To modify an existing word, use update_word(). Requires EUDIC_AUTH_TOKEN environment variable.'
• Return structured objects from tools instead of plain-text responses. Wrap results in JSON with typed fields (id, created_at, errors) so LLMs can extract data predictably and chain calls without parsing strings.