Both tools have minimal descriptions (under 50 chars) and lack parameter descriptions entirely. The 'search' tool accepts a free-form query string with no constraints, format guidance, or examples of valid input. The 'get' tool accepts a URL with no validation hints or error recovery guidance. Neither tool documents output schema structure, search returns a list of dicts with title/abstract/url, but this is inferred from code, not declared. No error handling, no pagination for search results (hardcoded to 50), no input validation. Tool names are acceptable (verb_noun pattern), but descriptions are too terse to guide LLM selection. Schemas are partially visible in code but not formally declared in tool registration.

• Expand search description to 100-150 chars: 'Search arXiv papers by title, author, or abstract. Returns up to 50 most recent results with title, abstract, and URL. Use natural language queries (e.g., "machine learning transformers") or author names.'
• Add parameter description for 'query': 'Free-form search string. Supports natural language (e.g., "neural networks"), author names (e.g., "Bengio"), or arXiv field codes (e.g., "cs.AI"). Searches title, abstract, and author fields.'
• Document output schema in tool description or return type annotation: 'Returns list of papers, each with: title (str), abstract (str), url (str, relative arXiv URL).'
• Add try-except blocks with actionable error messages: 'arXiv API unavailable. Retry in 30 seconds.' or 'Invalid URL format. Ensure URL starts with /abs/ or /pdf/.'
• Add parameter description for 'get' URL: 'Relative or absolute arXiv URL (e.g., "/abs/2301.12345" or "https://arxiv.org/abs/2301.12345"). Tool fetches full paper text via Jina reader.'
• Implement pagination: add optional 'start' and 'limit' parameters to search, document the total count in response, and explain how to fetch next page.