Loading ToolBench
Academic paper search MCP server for Cloudflare Workers (mcp-deploy compatible). Multi-platform academic paper discovery with 6 composable tools for searching, discovering, ranking, and analyzing research papers.
Paper Search MCP demonstrates solid tool design with well-structured schemas, clear naming conventions, and comprehensive parameter documentation. All 6 tools follow verb_noun naming (search_papers, discover_recent_papers, etc.). Input schemas are complete with type definitions and descriptions for all parameters. However, output schemas are not explicitly documented in the source code provided, only input schemas are visible. Error handling guidance is minimal; tools lack recovery hints for common failure modes. Tool descriptions are adequate (100-200 chars) but could be more prescriptive about when to use each tool vs. alternatives. The ranking system with presets and weight dials is well-designed but adds complexity that could confuse LLM selection logic.
Find noteworthy new research using age-adaptive scoring. Discovers fresh, high-quality papers published recently with configurable ranking dials.
ML recommendations from seed papers using Semantic Scholar embeddings. Finds related work via ML embeddings for a given paper.
Citation graph traversal (forward/backward). Walk the citation graph to find foundational work (references) or follow-up work (citations) for a given paper.
Look up a single paper by any ID type (DOI, S2 corpus ID, arXiv ID, PMID). Returns full details including TLDR, fields of study, and author metrics.
Deep quality scoring on specific papers. Re-ranks a list of papers using venue quality, citation metrics, freshness, and other signals with configurable ranking dials.
Output schemas not documented in source. Tool descriptions state what is returned (e.g., 'Returns full details including TLDR, fields of study, and author metrics') but formal output schema definitions are absent. LLMs cannot plan downstream tool calls without knowing the exact fields returned.
No error recovery guidance. Tools lack descriptions of failure modes and recovery steps. E.g., if a paper_id is invalid or a platform times out, what should the LLM do? No guidance provided.
Ranking preset complexity may confuse LLM selection. Four presets (relevance, balanced, discovery, impact) plus four weight dials create 16+ combinations. Tool descriptions do not clearly state when to use each preset or how weight dials interact with presets.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Core multi-query search with RRF fusion across multiple academic platforms. Supports decomposed queries, date filtering, journal filtering, and configurable ranking presets (relevance, balanced, discovery, impact).
Paper ID parameter accepts multiple formats (DOI, S2 corpus ID, arXiv ID, PMID, OpenAlex ID) but does not document resolution order or error handling if format is ambiguous. LLMs may pass invalid IDs without clear feedback.
No pagination or result limits documented for tools returning lists. search_papers and discover_recent_papers accept max_results but do not specify absolute caps or whether results are paginated. Large result sets could exhaust context windows.