Inspired by and based on alaliqing/claude-paper
Turn papers into a reusable research knowledge system
A customized Claude Code plugin for studying research papers with Chinese-first material generation, AlphaXiv-assisted understanding, a knowledge graph, a persistent knowledge base, and an interactive web reader.
Compared with the original claude-paper, this repo is focused on a deeper research workflow:
- Chinese-first study output: generated study materials default to Chinese unless the user explicitly asks for English.
- AlphaXiv integration: for arXiv papers, the workflow fetches
alphaxiv.mdand cross-checks it with parsed PDF content. - Cross-paper knowledge base: each studied paper appends a linked note to
~/claude-papers/paper.md. - Semantic tags and graph view: papers get 2-3 reusable tags, plus a graph view connecting papers by tags, authors, and related ideas.
- Editable web workspace: edit generated files, preview HTML, save snippets to notes, and run local demo code directly from the web UI.
- Multiple input modes: local PDFs, direct PDF URLs, arXiv URLs, multiple papers at once, or a directory of PDFs.
- Structured PDF parsing: extracts title, authors, abstract, sections, references, code links, and full text.
- AlphaXiv augmentation: supplements arXiv papers with machine-readable summaries from AlphaXiv.
- Study material generation: creates
README.md,summary.md,method.md,reflection.md, anduser.md. - Figure extraction: saves paper figures into an
images/directory for downstream study materials. - Code demos: generates runnable examples in
code/when a paper has an implementable method. - Interactive explorer: can generate a self-contained
index.htmlfor architecture-, system-, or experiment-heavy papers. - Knowledge base updates: appends a concise cross-paper note into the global
paper.md. - Rich web UI: library view, reader view, knowledge graph, global notes page, tag editing, read/star status, delete, edit, and run.
/plugin marketplace add https://github.com/Hoder-zyf/claude-paper
/plugin install claude-paper
/reload-pluginscd /path/to/claude-paper
claude --plugin-dir ./plugin- Node.js: 18+
- npm
- Claude Code with plugin support
- Python 3
- poppler-utils for figure extraction
- macOS:
brew install poppler - Ubuntu/Debian:
sudo apt-get install poppler-utils - Arch Linux:
sudo pacman -S poppler
- macOS:
On first run, the plugin installs its own Node dependencies and initializes ~/claude-papers/.
Examples:
Help me study ~/Downloads/attention-is-all-you-need.pdf
Help me study https://arxiv.org/pdf/1706.03762.pdf
Help me study https://arxiv.org/abs/1706.03762
Help me study these two papers: <url-1> <url-2>
Help me study all papers in ~/papers/
You can also invoke the bundled slash command directly:
/claude-paper:study /path/to/paper.pdfThe workflow will:
- Resolve inputs and download PDFs when needed.
- Parse the paper and save
meta.json. - Fetch
alphaxiv.mdfor arXiv papers when available. - Extract figures into
images/. - Read and update the global knowledge base.
- Generate study materials and optional code demos.
- Update
~/claude-papers/index.jsonand tag registry. - Launch the web UI.
/claude-paper:webuiThe viewer runs at http://localhost:5815.
Each paper lives under ~/claude-papers/papers/{paper-slug}/:
~/claude-papers/
├── index.json
├── paper.md
├── tags.json
└── papers/
└── {paper-slug}/
├── paper.pdf
├── meta.json
├── alphaxiv.md # optional, for arXiv papers
├── README.md
├── summary.md
├── method.md
├── reflection.md
├── user.md
├── index.html # optional interactive explorer
├── images/
│ └── ...
└── code/
└── ...
Global files:
~/claude-papers/index.json: searchable paper index used by the web UI.~/claude-papers/paper.md: cross-paper knowledge base.~/claude-papers/tags.json: canonical tag registry.
The Nuxt-based web app includes:
- Library: browse, search, filter, sort, star, mark as read, retag, and delete papers.
- Paper Reader: navigate all generated files for one paper.
- Knowledge Graph: visualize links by shared tags, shared authors, and related ideas.
- Knowledge Base: edit and review the global
paper.md. - In-browser editing: modify generated Markdown/code files.
- Run button: execute local
.py,.js,.ts, and.shdemo files. - Save to Notes: clip selected content into
user.md.
claude-paper/
├── .claude-plugin/
├── plugin/
│ ├── .claude-plugin/
│ ├── commands/
│ │ ├── study.md
│ │ └── webui.md
│ ├── hooks/
│ ├── skills/
│ │ └── study/
│ │ ├── SKILL.md
│ │ ├── alphaxiv-paper-lookup.md
│ │ └── scripts/
│ ├── src/
│ │ └── web/
│ └── package.json
├── README.md
└── README.zh-CN.md
Key pieces:
plugin/skills/study/SKILL.md: the end-to-end paper study workflow.plugin/skills/study/alphaxiv-paper-lookup.md: AlphaXiv lookup helper.plugin/commands/study.md: slash command entry for study.plugin/commands/webui.md: production web viewer launcher.plugin/src/web/: Nuxt UI and Nitro APIs.
node plugin/skills/study/scripts/parse-pdf.js /path/to/paper.pdfcd plugin/src/web
npm install
npm run devcd plugin/src/web
npm run buildcd /path/to/claude-paper
claude --plugin-dir ./pluginThen run:
/claude-paper:study /path/to/paper.pdfDefaults:
- Papers directory:
~/claude-papers/ - Web UI port:
5815 - Code execution timeout in web UI:
30s
The main workflow lives in plugin/skills/study/SKILL.md, so prompt behavior and output conventions can be adjusted there.
This project is licensed under the MIT License. See LICENSE.
- Built with Claude Code
- Inspired by and based on alaliqing/claude-paper
- Thanks to AlphaXiv for providing machine-readable paper overviews
- Web viewer built with Nuxt.js
- Math rendering powered by KaTeX
- PDF parsing powered by pdf-parse