The arXiv Research MCP server has significant definition quality gaps. Tool naming is generally adequate (verb-forward: search_, get_, clear_) but descriptions vary widely in completeness and clarity. Most critically, the server exposes 5 tools with incomplete schema documentation and missing parameter descriptions. The inferred tool definitions (from integrations/langchain_tool.py) lack explicit visible registration in source control, making it difficult to verify actual implementation. While some tools like search_arxiv_papers have reasonable parameter documentation, others (clear_cache, get_cache_stats, arxiv_cache_management) have minimal or generic descriptions. No output schemas are documented in the tool definitions, LLMs cannot know what fields to expect in responses. Error handling guidance is absent entirely. This is typical of community research tools but falls well short of production quality.

• Document output schema for all 5 tools. Specify what fields are returned (e.g., for search_arxiv_papers: [{id, title, authors, abstract, url, relevance_score, full_text?}]). Include field types and whether fields are always present.
• Add confirmation/dry-run pattern for destructive tools. For clear_cache and arxiv_cache_management 'clear' action, implement a confirm_clear_cache(confirmed: bool) pattern or return a dry-run result first ('This will delete 150 cached entries. Proceed? Yes/No').
• Add error handling guidance. Example: 'search_arxiv_papers may fail if: (1) arXiv API is unreachable, wait 30s and retry; (2) query is empty, provide a non-empty search term; (3) years_back exceeds arXiv history, use years_back ≤ 30.'
• Consolidate search_arxiv_papers and arxiv_research into a single canonical search tool. If intentional duplicates (for compatibility), explain in descriptions when each should be used (e.g., 'arxiv_research is the preferred tool; search_arxiv_papers is a legacy interface').
• Add parameter constraints to descriptions: 'max_results: 1 - 500 (default 10); years_back: 0 - 30 (default 4); query: non-empty string, max 256 chars.'
• Verify tool definitions in scripts/run_server_http.py. Ensure tool registration explicitly declares names, descriptions, and input schemas. Confirm output schema is documented.
• Add 'include_full_text' parameter safety guidance: 'When true, full_text may be very large (100KB+). Use sparingly to avoid context window exhaustion. Consider false for initial discovery, true only after narrowing results.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry