10 tools with consistent naming (verb_noun pattern: search, get, list) and reasonable descriptions (50-150 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. No error handling guidance, no pagination support despite list tools, and no tool annotations (readOnlyHint). Parameter descriptions are adequate but lack constraints (e.g., 'limit' accepts 1-20 but no validation guidance). Missing idempotency and recovery patterns.

• Document output schemas for all tools. Example: arxiv_search returns {papers: [{id, title, authors, published, summary}], total_count}. This enables agents to extract and chain results.
• Add error recovery hints to descriptions. E.g., arxiv_get_paper: 'If paper not found, try arxiv_search() with keywords from the title.'
• Annotate all tools with readOnlyHint=true in FastMCP tool decorator. This signals to agents that these tools are safe to retry.
• Add pagination to list tools: arxiv_recent and countries_by_region should accept offset/limit and return total_count. Cap results at 20 by default.
• Validate parameter constraints (limit 1-20, category enum values) and return actionable errors: 'Invalid limit: 25. Must be 1-20.' instead of generic 500 errors.