PubMed Weekly Digest
didiowen/pubmed-weekly-digest
Digest semanal automatizado de literatura médica do PubMed com TL;DR e Hot Takes
Ver no GitHubSkill do Claude Code que gera um resumo semanal de artigos de journals indexados no PubMed dos últimos 7 dias, anotando cada um com TL;DR e comentários opinados. Salva o resultado em arquivo Markdown com suporte a múltiplos idiomas e integração com Gmail para newsletters.
Instalação
npx convez-ai didiowen/pubmed-weekly-digest1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Monitorar publicações recentes em journals médicos específicos
- Gerar newsletters semanais de pesquisa para equipes clínicas
- Acompanhar atualizações em áreas específicas da medicina com resumos opinados
Veredito

Ferramenta bem estruturada para cientistas e profissionais médicos que querem manter-se atualizados. Oferece três níveis de complexidade (básico, nuvem e local) e integração com Gmail, mas requer familiaridade com Claude Code e configuração de MCPs.
README
A tradução para o português deste README ainda está na fila.
Pubmed Weekly Digest
A Claude Code skill that builds a weekly digest of articles published in the last 7 days across a configurable list of PubMed-indexed journals, annotates each with a TL;DR and a one-line Hot Take (positive or snarky), and writes the result to a local Markdown file. Output is in English by default and the language is one prompt-edit away from anything else.
Folder layout
pubmed-weekly-digest/
├── SKILL.md # quickstart: simple, English, PubMed only
├── SKILL_cloud.md # full-featured: Chinese TL;DR + comment, Gmail newsletter, PMID dedup
├── SKILL_local.md # local-only: same as cloud + redirect DOI resolution + institutional browser
├── README.md # this file
├── README.zh-TW.md # Chinese version
└── LICENSE
No bundled scripts — the entire skill is prompt-driven.
Which skill file to use?
| File | When to use | Requirements |
|---|---|---|
SKILL.md | Quickstart — English output, any journal list, no Gmail needed | PubMed MCP |
SKILL_cloud.md | Full-featured — Traditional Chinese TL;DR + 嘻嘻/不嘻嘻 comment, NEJM Gmail newsletter (ID + Hema/Onc monthly), PMID dedup, OA abstract fallback; works on cloud and local | PubMed MCP + Gmail MCP |
SKILL_local.md | Same as SKILL_cloud.md, plus redirect-based DOI resolution for NEJM newsletter tracking links and institutional browser fallback for paywalled abstracts | PubMed MCP + Gmail MCP + local Claude-in-Chrome |
All three share the same journal list, ISO week-label logic, and CrossRef abstract refill. The full-featured variants add a Gmail newsletter layer (NEJM Infectious Disease + Hematology/Oncology monthly updates) and a smarter abstract pipeline.
SKILL.mdis intentionally kept as the simple, language-agnostic entry point. For a production setup with all the features, start withSKILL_cloud.mdinstead.
Dependencies
- Claude Code (with the Skill system) —
https://claude.com/claude-code - PubMed MCP server configured in your Claude Code MCP settings
- Gmail MCP server (only for
SKILL_cloud.md/SKILL_local.md) — needed to pull the NEJM monthly newsletters. If you subscribed to NEJM newsletters with a Gmail account, the Gmail MCP works out of the box. If you used a different address (e.g. ProtonMail), set up forwarding to Gmail first.
Setup
- Drop this folder into the skills directory of any project:
<project>/.claude/skills/pubmed-weekly-digest/. - Confirm the PubMed MCP server is configured (and Gmail MCP if using the full-featured variants).
- Open the skill file you chose and edit the Configuration block at the top:
OUTPUT_DIR: where to write the digest (defaultoutput/inside this skill folder)JOURNALS: the list in Step 1 (only inSKILL.md)CROSSREF_MAILTO: any valid email — used as the CrossRef polite-pool identifierSKILL_DIR: path to this skill folder, used for the PMID dedup cache (SKILL_cloud.md/SKILL_local.mdonly)
- Trigger from Claude Code: ask it to "run the weekly journal digest" or "跑週報", or invoke the skill by name.
Customising the journal list
This is the main thing you'll want to change.
Where the list lives
SKILL.md → Step 1 — PubMed search has a table of journal queries. Each row becomes one parallel mcp__PubMed__search_articles call. Replace rows to follow different journals, add rows for more journals, or delete rows you don't care about. Also update the section list in Step 5 — each journal gets its own ## {Journal name} block in the Markdown template.
Finding the right journal abbreviation
PubMed's [Journal] field expects the NLM Title Abbreviation, not the full name. Three ways to find it:
- Search the journal on PubMed → open any recent article → the citation line shows the abbreviation, e.g.
Clin Infect Dis. 2026 Apr 30;82(8):.... - Browse https://www.ncbi.nlm.nih.gov/nlmcatalog/journals → look up by full title → the "NLM Title Abbreviation" field is what to paste into
[Journal]. - Use
mcp__PubMed__search_articlesonce with the full name in your test query — PubMed normalises it and the response shows the abbreviation it actually used.
Common abbreviations:
| Full title | [Journal] value |
|---|---|
| Clinical Infectious Diseases | Clin Infect Dis |
| Emerging Infectious Diseases | Emerg Infect Dis |
| The New England Journal of Medicine | N Engl J Med |
| The Lancet | Lancet |
| The Lancet Infectious Diseases | Lancet Infect Dis |
| JAMA | JAMA |
| Journal of Clinical Oncology | J Clin Oncol |
| Nature Medicine | Nat Med |
| The Journal of Infectious Diseases | J Infect Dis |
| Blood | Blood |
| BMJ | BMJ |
| Annals of Internal Medicine | Ann Intern Med |
Picking max_results
The MCP wrapper caps at 200 per query. Rough sizing for a 7-day window:
- High-volume specialty journals (CID, EID, Blood):
50is usually plenty. - Weekly bulletins (MMWR, Lancet):
30covers one issue. - High-impact generals (NEJM, JAMA):
50is a safe upper bound; one issue is usually under 30. - Monthly journals (most subspecialty journals):
30is more than enough.
Two PubMed MCP pitfalls to remember
- Wildcards
*are not supported. Usemycobacteriuminstead ofmycobacteri*, or expand explicitly:mycobacterium OR tuberculosis OR NTM. - The boolean-operator cap is 20. A single query with more than 20
OR/ANDoperators fails withINVALID_PARAMETERS. Split into two parallel queries if you hit this.
Adapting the output style
The annotations in Step 4 are written in English by default, with the one-line comment framed as a Hot Take. Both choices are pure prompting — adapt as you like:
- Language: edit the instructions in Step 4 to specify your preferred language (e.g. Traditional Chinese, Spanish, Japanese). Claude translates abstracts on the fly; no other part of the skill needs changing.
- Tone label: rename "Hot Take" to whatever framing fits — "key takeaway", "clinical implication", "bottom line", etc.
- Length / depth: the default ask is 1–2 sentences. Loosen it to a paragraph if you want more detail.
License
MIT — see LICENSE.