PubMed Skills for Claude Code

English | 中文

---

English

Claude Code skills that let Claude interact with PubMed through Chrome DevTools MCP.

Search papers, view detailed metadata, access full text, and export to Zotero with PDF attachments — all from the Claude Code CLI.

Prerequisites

• Claude Code CLI installed
• Chrome browser
• Zotero desktop app (optional, for export)
• Python 3 (optional, for Zotero push script)

Skills

Agent

pm-researcher — orchestrates all 6 skills. Handles PubMed API interaction, tab management, and multi-step workflows like “search → filter → export to Zotero”.

Installation

1. Install Chrome DevTools MCP server

claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest

2. Install PubMed skills

git clone https://github.com/cookjohn/pm-skills.git
cd pm-skills
cp -r skills/ agents/ .claude/

Or add to an existing project:

git clone https://github.com/cookjohn/pm-skills.git /tmp/pm-skills
cp -r /tmp/pm-skills/skills/ your-project/.claude/skills/
cp -r /tmp/pm-skills/agents/ your-project/.claude/agents/

3. Start Chrome with remote debugging

# Windows
"C:\Program Files\Google\Chrome\Application\chrome.exe" --remote-debugging-port=9222

# macOS
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222

# Linux
google-chrome --remote-debugging-port=9222

4. Launch Claude Code

claude

Skills and agent are picked up automatically. Try /pm-search CRISPR to verify.

How It Works

All skills use PubMed’s E-utilities API (esearch, esummary, efetch) via evaluate_script calls through Chrome DevTools MCP. The export skill uses a 3-step flow to attach PDFs to Zotero:

1. saveItems — push paper metadata to Zotero via Connector API
2. download PDF — fetch PDF from PMC/publisher via Python
3. saveAttachment — upload PDF binary to Zotero, linked to the parent item

Key design choices:

• PubMed API over DOM scraping — structured JSON responses, no fragile selectors
• Deterministic session IDs — content-hash based, prevents duplicate imports
• PDF attachment via saveAttachment — Zotero 7.x ignores the attachments field in saveItems, so PDFs are uploaded separately

Project Structure

skills/
├── pm-search/SKILL.md              # Basic keyword search
├── pm-advanced-search/SKILL.md     # Filtered search (author, journal, date, MeSH)
├── pm-paper-detail/SKILL.md        # Paper metadata extraction
├── pm-fulltext/SKILL.md            # Full-text access links
├── pm-navigate-pages/SKILL.md      # Pagination
└── pm-export/                      # Zotero export with PDF
    ├── SKILL.md
    └── scripts/
        └── push_to_zotero.py       # Zotero Connector API client
agents/
└── pm-researcher.md                # Agent: orchestrates all skills

---

中文

让 Claude Code 通过 Chrome DevTools MCP 操作 PubMed 的技能集。

支持论文检索、详细信息查看、全文获取、导出到 Zotero（含 PDF 附件）等功能，全部在 Claude Code 命令行中完成。

前置要求

• 已安装 Claude Code CLI
• Chrome 浏览器
• Zotero 桌面端（可选，用于导出）
• Python 3（可选，用于 Zotero 推送脚本）

技能列表

智能体

pm-researcher — 统一调度全部 6 个技能。处理 PubMed API 交互、标签页管理，支持"检索 → 筛选 → 导出到 Zotero"等复合工作流。

安装方法

1. 安装 Chrome DevTools MCP 服务器

claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest

2. 安装 PubMed 技能

git clone https://github.com/cookjohn/pm-skills.git
cd pm-skills
cp -r skills/ agents/ .claude/

添加到已有项目：

git clone https://github.com/cookjohn/pm-skills.git /tmp/pm-skills
cp -r /tmp/pm-skills/skills/ your-project/.claude/skills/
cp -r /tmp/pm-skills/agents/ your-project/.claude/agents/

3. 启动 Chrome 远程调试

# Windows
"C:\Program Files\Google\Chrome\Application\chrome.exe" --remote-debugging-port=9222

# macOS
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222

# Linux
google-chrome --remote-debugging-port=9222

4. 启动 Claude Code

claude

技能和智能体会自动加载。输入 /pm-search CRISPR 验证是否正常。

工作原理

所有技能通过 Chrome DevTools MCP 的 evaluate_script 调用 PubMed E-utilities API（esearch、esummary、efetch），获取结构化 JSON 数据。导出技能使用三步流程将 PDF 附件添加到 Zotero：

1. saveItems — 通过 Connector API 推送论文元数据到 Zotero
2. 下载 PDF — 通过 Python 从 PMC/出版商获取 PDF
3. saveAttachment — 上传 PDF 二进制文件到 Zotero，关联到父条目

核心设计：

• PubMed API 优于 DOM 抓取 — 结构化 JSON 响应，无脆弱的 CSS 选择器
• 确定性会话 ID — 基于内容哈希，防止重复导入
• 通过 saveAttachment 附加 PDF — Zotero 7.x 忽略 saveItems 中的 attachments 字段，PDF 需单独上传

---

License

MIT