What problem does it solve? Finding relevant academic papers across multiple databases is slow and error-prone: queries must be reformulated per database, results overlap, and metadata is inconsistent. This Skill automates verified literature retrieval through MCP connectors and bundled search scripts, returning a deduplicated, machine-readable source package ready for downstream evidence extraction. ## Core Features & Use Cases - Multi-database search: Queries arXiv, PubMed, and CrossRef with domain-aware routing (biomedicine, chemistry, materials, finance, computer science), date filters, arXiv categories, and PubMed MeSH terms. - Query planning and variants: Extracts compact core keywords, expands acronyms (e.g., GNN to graph neural network), and generates up to three query variants when initial results fall below threshold. - Deduplication and coverage checks: Merges results by DOI, URL, and title similarity (default 0.85), scores relevance, documents coverage gaps, and returns a SUFFICIENT, PARTIAL, or INSUFFICIENT verdict. - Use Case: Given the topic "graph neural networks for molecular property prediction", the Skill searches arXiv and PubMed, filters off-topic results, deduplicates overlapping records, and writes a literature_sources.json file that an evidence-extraction step can consume directly. ## Quick Start Search the literature for recent papers on transformer attention mechanisms in NLP and produce a deduplicated JSON source package with a coverage assessment.
