PubMed Export to Zotero

Export PubMed paper citation data and push to Zotero desktop via local Connector API, or generate MEDLINE/RIS format.

Arguments

$ARGUMENTS contains one or more PMIDs (space-separated), e.g.:

• 35641793 — single paper
• 35641793 27741350 35362092 — batch export

Steps

Step 1: Fetch paper metadata (evaluate_script)

Use mcp__chrome-devtools__evaluate_script on any PubMed page. Replace PMID_LIST with the actual PMIDs:

async () => {
  const pmids = "PMID_LIST".split(/[\s,]+/).filter(Boolean);

  // Batch esummary for basic metadata
  const sumResp = await fetch(
    `https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi?db=pubmed&id=${pmids.join(',')}&retmode=json`
  );
  const sumData = await sumResp.json();

  // efetch XML for full records (abstracts, keywords)
  const fetchResp = await fetch(
    `https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=${pmids.join(',')}&retmode=xml`
  );
  const xml = await fetchResp.text();
  const xmlDoc = new DOMParser().parseFromString(xml, 'text/xml');

  const papers = pmids.map(pmid => {
    const r = sumData.result?.[pmid] || {};
    const doi = (r.articleids || []).find(a => a.idtype === 'doi')?.value || '';

    // Find this paper's article in the XML
    const articles = xmlDoc.querySelectorAll('PubmedArticle');
    let abstract = '', keywords = [], meshTerms = [], pubTypes = [];
    let issn = '', pmcid = '';
    for (const art of articles) {
      if (art.querySelector('PMID')?.textContent === pmid) {
        abstract = Array.from(art.querySelectorAll('AbstractText'))
          .map(el => el.textContent).join(' ');
        keywords = Array.from(art.querySelectorAll('Keyword'))
          .map(k => k.textContent);
        meshTerms = Array.from(art.querySelectorAll('MeshHeading DescriptorName'))
          .map(d => d.textContent);
        pubTypes = Array.from(art.querySelectorAll('PublicationType'))
          .map(p => p.textContent);
        issn = art.querySelector('ISSN')?.textContent || '';
        pmcid = art.querySelector('ArticleId[IdType="pmc"]')?.textContent || '';
        break;
      }
    }

    return {
      pmid,
      title: r.title || '',
      authors: (r.authors || []).map(a => ({
        lastName: a.name?.split(' ')[0] || '',
        firstName: a.name?.split(' ').slice(1).join(' ') || '',
      })),
      journal: r.fulljournalname || '',
      journalAbbr: r.source || '',
      pubdate: r.pubdate || '',
      volume: r.volume || '',
      issue: r.issue || '',
      pages: r.pages || '',
      doi,
      issn,
      pmcid,
      abstract,
      keywords,
      pubtype: pubTypes,
      language: (r.lang || ['eng'])[0] === 'eng' ? 'en' : (r.lang || ['eng'])[0]
    };
  });

  return papers;
}

Step 2: Push to Zotero

Save the returned JSON array to a temp file, then run the Python script:

python "e:/pm-skills/.claude/skills/pm-export/scripts/push_to_zotero.py" /tmp/pm_papers.json

The Python script runs a 3-step flow:

1. saveItems — saves metadata to Zotero (POST /connector/saveItems). Note: Zotero 7.x ignores the attachments field in saveItems.
2. download_pdf — downloads PDF binary via Python urllib (if pdfUrl or pmcid is available)
3. saveAttachment — uploads PDF binary to Zotero (POST /connector/saveAttachment), linked to the parent item via parentItemID in the X-Metadata header

Other details:

• Builds Zotero-compatible item JSON (itemType: journalArticle, libraryCatalog: PubMed)
• Parses PubMed author format into Zotero's {firstName, lastName} structure
• Sets DOI, PMID (in extra field), URL to PubMed page
• Uses deterministic session ID for idempotency (same papers → same session → no duplicates)
• PDF download may fail for some publishers (403, JS-redirect pages); these are reported as "PDF skip"

Step 3: Report

Single paper:

Exported to Zotero:
  Title: {title}
  Authors: {authors}
  Journal: {journal} ({pubdate})
  DOI: {doi}
  PMID: {pmid}

Batch:

Exported {count} papers to Zotero:
  1. {title1} ({journal1}, {pubdate1})
  2. {title2} ({journal2}, {pubdate2})
  ...

MEDLINE Export (alternative)

If the user wants a file instead of Zotero, use efetch to get MEDLINE format directly:

async () => {
  const pmids = "PMID_LIST";
  const resp = await fetch(
    `https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=${pmids}&rettype=medline&retmode=text`
  );
  return await resp.text();
}

Save the returned text as a .nbib file. This can be imported into Zotero, EndNote, or any reference manager.

Zotero Troubleshooting

• Zotero not running: Script returns "Zotero is not running". User must start Zotero desktop.
• Read-only library: Script returns "Target library is read-only". User must select a writable collection.
• Already saved: Script returns 409, treated as success (idempotent).

Notes

• This skill uses 2 tool calls: evaluate_script + bash python
• Batch export works for any number of PMIDs in a single call
• The Python script is at e:/pm-skills/.claude/skills/pm-export/scripts/push_to_zotero.py
• Zotero must be running with the Connector API on localhost:23119