Two tools with complete schemas and descriptions. Naming follows verb_noun pattern (search_papers, get_paper). Descriptions are detailed (150-200 chars) and explain use cases well. Input schemas are properly typed with enums for sort_by/sort_order. However, output schemas are NOT documented, callers cannot see what fields to expect from results. No error handling guidance. No parameter constraints (max_results accepts 1-100 but lacks minumum/maximum in schema). Missing idempotency hints and recovery guidance.
Fetch a single arXiv paper by its ID (e.g., "2310.06825", "2310.06825v2", or legacy "cs.CL/0301001"). Returns full metadata: title, authors, abstract, categories, DOI (if linked), PDF URL.
Search arXiv preprints. Plain text searches all fields; use prefixes for targeted queries: au:hinton (author), ti:transformer (title), abs:diffusion (abstract), cat:cs.AI (category), all:quantum (any field). Combine with AND/OR/ANDNOT, e.g., "ti:transformer AND cat:cs.LG". Returns id, title, authors, abstract, categories, published date, PDF URL.
Output schemas not documented. Callers cannot see what fields search_papers and get_paper return. LLMs cannot plan downstream operations or extract required data.
max_results parameter lacks minimum/maximum constraints in schema. Description says '1-100' but schema has no minumum/maximum properties. LLMs cannot validate bounds and may pass invalid values.
No error handling guidance. If arXiv API fails or arxiv_id is invalid, no recovery hints provided. LLMs cannot self-correct or know what to try next.
No idempotency hints. get_paper appears idempotent (read-only) but this is not declared. search_papers with same query should return same results, but no guarantee stated.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
search_papers returns 'total_results' and 'results' but no pagination support. Large result sets could blow context window. No limit enforcement or next_cursor documented.