Loading ToolBench
MCP server for data scientists - search ArXiv papers, find code, get citations, track SOTA benchmarks
Server has 10 tools with complete naming (verb_noun pattern) and descriptions (avg ~120 chars). All tools are READ_ONLY, reducing risk. However, parameter descriptions are sparse or missing for many inputs. Output schemas are not explicitly documented in the code, responses are formatted as markdown strings rather than structured JSON. Error handling is minimal; no recovery guidance or actionable error messages visible. Missing tool annotations (readOnlyHint, idempotentHint). Composition is sound (each tool has one responsibility), but chaining support is weak, no IDs returned to enable downstream calls.
Find code repositories and implementations associated with a research paper from Papers with Code.
Find papers similar to a given paper based on content and citations.
Find Hugging Face models and datasets related to a research paper or topic.
Get citation information for a paper including citing papers, citation count, and citation context.
Fetch detailed information about a specific ArXiv paper including full metadata, citations, references, and code repositories.
Get the reference list (papers cited by) a specific paper.
Get state-of-the-art benchmark results and leaderboards for a research task or dataset.
Get trending and recently published papers in specified categories with high citation velocity.
Output schemas not documented. Tools return markdown-formatted strings rather than structured JSON. LLMs cannot reliably extract fields for downstream tool calls (e.g., arxiv_id from search results to pass to get_paper_details).
Parameter descriptions incomplete or missing. 'limit' and 'max_results' parameters lack min/max bounds. 'sort_by' enum values documented in description but not as formal enum constraint. 'category' accepts free-form strings with no validation guidance.
No error handling or recovery guidance visible in code. External API calls (ArXiv, Semantic Scholar, Papers with Code) have no timeout handling, retry logic, or user-actionable error messages. Failed calls will return raw exceptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Search ArXiv papers by query, category, and sorting options. Returns structured paper metadata including title, authors, abstract, citations, and links.
Search for papers by a specific author across ArXiv.
No tool annotations. All tools are read-only but lack readOnlyHint annotation. This prevents clients from optimizing caching, retry logic, or safety checks.
Pagination support incomplete. search_arxiv and search_papers_by_author accept 'start' and 'max_results' but do not return total_count or next_cursor. Agents cannot reliably iterate through large result sets.