Introduction

In the plugin ecosystem of DeepSeek Harness (DSH), search tools oriented toward computer science and interdisciplinary domains, such as arXiv, dblp, and OpenAlex, are already available, but a dedicated plugin for PubMed, the authoritative biomedical database, is still absent. The dsh-pubmed plugin fills this functional gap. It provides a set of advanced search tools that directly connect to the public NCBI E-utilities interfaces and support end-to-end operations, from literature search to author profiling, institutional statistics, and retraction checks, without needing to apply for an API key.

Core Features

This plugin provides seven basic tools covering the main requirements for biomedical information retrieval and investigation:

  • Literature Search: Supports field tags such as [Author] and [Affiliation], and returns the number of hits and a PMID list.
  • Batch Retrieval of Literature Metadata: Through the pubmed_fetch tool, you can retrieve titles, journals, years, authors, and DOIs in batches.
  • Author Publication Investigation: By entering a full name (in pinyin), you can query the author’s publication volume and institutional distribution.
  • Same-Name Author Disambiguation: Uses affiliation information to distinguish authors with the same name and resolves search ambiguity.
  • Institutional Publication Statistics: Counts a specific institution’s publications and their year-by-year distribution.
  • Supervisor-Student Co-Authorship Matching: Finds records of a specific supervisor and student sharing authorship.
  • Retraction Check: Searches for retraction records related to a specific person or topic.

Installation and Activation

Installing the plugin requires the DSH CLI. Replace <repository URL or npm package name> with the actual link or package name:

dsh plugin --profile web add <仓库地址或 npm 包名>

For local development environments, you can use the file protocol to point directly to a local path:

dsh plugin --profile web add file:/path/to/dsh-pubmed

After installation, restart the DSH process. The seven tools described above will appear in the tool list.

Typical Usage

The tools are triggered by natural language instructions, and the Agent will automatically call the corresponding underlying tools.

  1. Query an Author’s Publication Volume and Affiliations:
    Input:
    帮我查一下作者 "Zhang Wei" 在 PubMed 发了多少篇

The Agent will call `pubmed_author` and `pubmed_disambiguate` in sequence to process the request.
  1. Institutional Statistics:
    Input:
    查一下四川大学 2026 年发了多少篇 PubMed 论文，按年份分布

The Agent will call the `pubmed_institution` tool.
  1. Retraction Check:
    Input:
    这个作者有没有被撤稿的记录？

The Agent will call the `pubmed_retraction` tool.

Use Cases and Notes

  • Chinese Search Rules: PubMed uses a pinyin indexing rule of “surname first, given name second” for Chinese authors. When searching Chinese authors, use the full pinyin name (for example, Zhang Wei), not Wei Zhang.
  • API Rate Limiting: The NCBI interface requires no more than 3 requests per second. The plugin has a built-in 350ms throttling mechanism. Large-scale searches may be relatively slow, which is the necessary cost of complying with NCBI rules.
  • Index Precision: Author full-name indexing relies on PubMed’s full ForeName spelling. Using only abbreviations (for example, Shen H) may lead to missed results. pubmed_disambiguate avoids this issue by using efetch to obtain full names and affiliations for precise matching.
  • Institutional Filtering: Institution filtering is word-level search rather than substring search. Minor differences in how an institution name is written, such as spaces or punctuation, may lead to missed results. It is recommended to first use a broad query and then perform precise verification.
  • Security and Dependencies: The plugin does not read DeepSeek credentials, does not make any third-party network requests, has no runtime dependencies, and only uses the built-in fetch in Node.js.

dsh-pubmed enables in-depth exploration of PubMed through purely public interfaces, making it suitable for developers who need to process biomedical literature, conduct academic investigations, or generate institutional statistics.

Catalog page | Source repository