dsh-pubmed: An Entity-Aware Literature Search & Knowledge Graph Plugin for DeepSeek Harness

What Does It Do?

dsh-pubmed is a DSH plugin for DeepSeek Harness that integrates PubMed, Europe PMC, PubTator3, and Semantic Scholar into 25 native model tools, without requiring an MCP client or paid proxy. It addresses common literature-search pain points such as synonym misses, noise from keyword co-occurrence, broken evidence chains, missing citation data, and fragmented multi-platform workflows, turning 80% of mechanical database screening time into 20% of high-quality reading time. Its core capabilities include entity-aware search, a personal literature knowledge graph, cross-source unified search, and auditable evidence.

Core Features

• Entity-aware search: normalizes free text to authoritative concept IDs (e.g., metformin → MESH:D008687) via PubTator3, and supports relation-style queries that directly return articles supporting a given relationship, immune to synonym noise.
• Personal literature knowledge graph: automatically merges each fetched article into the session graph, with keyword nodes, concept nodes (deduplicated across articles by authoritative ID), curated relation edges (with evidence PMIDs), and heuristic relation edges; can be persisted to a local JSON file.
• Cross-source unified search: one query hits PubMed, Europe PMC, and OpenAlex simultaneously, deduplicating and merging by DOI/PMID/title, ranking multi-source hits first, and merging citation counts; optionally includes Semantic Scholar.
• Citation and recommendation tools: Semantic Scholar tools provide citation counts, paper recommendations, and exact title matching, using the official free API with no key required.
• Full-text and metadata retrieval: fetches structured articles by PMID (authors, abstract, MeSH, grants, DOI, PMCID), plus MeSH and ID tools.

How to Use This Plugin?

After enabling and restarting DSH, pubmed_* tools appear in all sessions and work with zero configuration. Start with a self-check: pubmed_spell_check({ query: 'microbiom' }) should return corrected: "microbiome". For a broad scan, use pubmed_search_papers (cross-source unified search) first; for specific biological entities or drug-disease relations, use pubmed_pubtator_search; for field-limited or date-range queries, use pubmed_search_articles. Each pubmed_fetch_articles call automatically merges into the session graph, and once satisfied, run pubmed_graph_commit to persist it locally.