16 tools with consistent verb-noun naming (search_, get_, paper_*, author_*). All tools have descriptions (avg ~80 chars, within baseline 34-392). Input schemas present with types and descriptions for all parameters. However, output schemas are NOT documented, critical gap for LLM planning. Error handling returns structured responses but lacks recovery guidance (e.g., 'try search_authors() first'). No tool annotations (readOnlyHint, destructiveHint). Parameter descriptions are adequate but lack format constraints (e.g., no enum for publication_types, no regex for IDs). Pagination support present (offset/limit) but no total_count or next_cursor documented in output. All tools are READ_ONLY, reducing risk, but composition chains are not optimized, e.g., paper_details returns paper_id but downstream tools don't clearly document what IDs they accept.

• Document output schemas for all tools. Specify which fields are returned (e.g., author_id, paper_id, title, citation_count) so LLMs can chain tools and extract IDs for downstream calls.
• Add tool annotations: set readOnlyHint=true for all 16 tools to explicitly declare safety and enable LLM optimizations for read-only operations.
• Convert free-form string parameters to enums. Define valid values for publication_types (e.g., ['Journal', 'Conference', 'Preprint']), fields_of_study, venue, sort, and from_pool in the schema.
• Enhance error messages with recovery guidance. When a lookup fails, suggest the next tool: 'Author not found. Try author_search(query="partial name") to find the author ID.'
• Document pagination behavior in output schema. Specify total_count (integer) and next_cursor (string, optional) so LLMs know when to fetch more results and how.