Loading ToolBench
MCP server for scientific article research from ArXiv and Google Scholar
Server has 9 tools with complete schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (search_*, download_*, read_*, list_*, cleanup_*). Descriptions are present but vary in depth, some are generic ('Search for scientific papers on Google Scholar') while others provide context ('optimized to handle common topics'). All tools have input schemas with types and descriptions. However, output schemas are not documented, parameter descriptions lack format constraints and ranges, and error handling is minimal. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Missing pagination guidance for search results. Overall: functional but below production baseline.
Clean up all downloaded PDF files
Download a PDF and attempt to read its text content
Download a PDF from a URL and save it locally
List all downloaded PDF files
Read text content from a downloaded PDF file
Search for scientific papers on ArXiv. The search is optimized to handle common topics like "quantum computing", "machine learning", "artificial intelligence", etc. and will automatically use appropriate ArXiv categories for better results. You can also use ArXiv syntax directly (cat:, ti:, abs:, au:).
Search ArXiv papers by specific category using exact category codes.
Output schemas not documented. LLMs cannot plan downstream calls or extract fields without knowing what search_arxiv, search_google_scholar, and read_pdf_text return. Baseline: 100% of A+ tools document return types.
Tool annotations missing. download_pdf, download_and_read_pdf, and cleanup_downloads have clear risk levels (WRITE, DESTRUCTIVE) but lack readOnlyHint/destructiveHint annotations. Agents cannot distinguish safe reads from state-modifying calls.
Pagination not addressed. search_arxiv, search_google_scholar, and search_both_sources accept maxResults but do not document offset/cursor handling or total count. Large result sets risk context window exhaustion.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Search for scientific papers on both ArXiv and Google Scholar
Search for scientific papers on Google Scholar
Parameter descriptions lack format constraints. 'pdfUrl' and 'filepath' have no validation hints (URL format, path restrictions). 'startPage' and 'endPage' lack min/max bounds. LLMs will pass invalid values without guidance.
Error handling not documented. No guidance on what errors tools return, whether they are retryable, or what the LLM should do next. E.g., if download_pdf fails due to network timeout vs. invalid URL, the agent has no recovery path.