The server defines 6 tools with complete input schemas and descriptions visible in src/index.ts. Naming follows verb_noun patterns (get_*, search_*, clean_*). Descriptions are present and between 10-200 characters, but lack operational guidance and context about when to use each tool vs alternatives. Parameter descriptions exist but are generic. Output schemas are not documented, LLMs cannot predict what fields are returned. The tools are READ_ONLY dominated (5 of 6), with minimal error handling guidance. No input validation constraints (enums, ranges) beyond defaults. The WeReadApi integration is opaque, actual return structures are invisible in the source provided.

• Document output schemas for all tools. Add a 'schema' block to each tool definition showing the structure of returned data. E.g., 'Returns: { books: [{bookId, title, author, progress, ...}], totalBooks, categories: [{name, count}] }'.
• Add min/max bounds to numeric parameters. E.g., max_results: 'integer (1 - 100, default 5)', count: 'integer (1 - 50, default 10)'. Include in both the schema 'minimum'/'maximum' fields AND the description text so LLMs see it.
• Use enums for constrained parameters. For highlight_style, define: 'enum (1|2|3|null, default null) for highlight types, or null for all styles'.
• Expand tool descriptions to include operational context. E.g., search_books: 'Use to find specific books by title/author. For a complete list of all books with stats, call get_bookshelf instead. Returns up to max_results matches; use fuzzy matching by default for typo tolerance.'
• Add recovery guidance in descriptions. E.g., get_book_notes_and_highlights: 'Returns highlights and notes for a book. If book_id is invalid, search_books first to find the correct ID.'
• Document pagination pattern. E.g., get_book_best_reviews: 'Paginated results. Use max_idx and synckey to fetch additional pages. Returns count and next_synckey (if more results available) so agents can iterate.'
• Implement input validation and error messages with actionable hints. When a lookup fails, return available alternatives: 'Book ID "invalid_id" not found. Did you mean: [list of similar books]?'
• Move WeReadApi into the visible source and document secret injection pattern. Verify that API tokens/credentials are injected server-side via environment variables, not passed as tool parameters. Audit that responses strip sensitive fields (auth tokens, session IDs).