Papers Research

📚 papers-skill

🔬 Claude Code skill for searching 200M+ academic papers — search, cite, download arXiv PDFs, extract text. Zero config, no MCP server, no API keys.

papers-skill is the Skill-mode port of the papers-mcp MCP server. Same features, shipped as a self-contained Claude Code plugin: one Python script + one SKILL.md orchestrating it.

Both projects are maintained by the same author. Pick MCP if you want a long-running server; pick this Skill if you want a zero-config plugin install.

---

Features

Subcommand	What it does
search	Semantic Scholar broad search (200M+ papers, with citation counts)
detail	Full metadata, TL;DR, references — auto-detects DOI / arXiv / S2 ID
citations	Papers that cite a given work
arxiv	arXiv preprint search
download	Fetch arXiv PDF locally
read	Extract PDF text via PyMuPDF

All APIs are free and require no authentication.

---

Install

Option 1 — As a Claude Code plugin (recommended)

/plugin marketplace add xwmxcz/papers-skill
/plugin install papers-skill@xwmxcz-plugins

Then in any new conversation, ask Claude something like:

帮我搜一下 retrieval augmented generation 相关的论文

Claude auto-loads the skill and dispatches the right subcommand.

Option 2 — Manual install (no plugin system)

If you don't want to use the plugin marketplace, you can drop the skill directory into Claude Code's user-skills folder directly:

# macOS / Linux
git clone https://github.com/xwmxcz/papers-skill ~/.claude/skills/papers-research-tmp
cp -r ~/.claude/skills/papers-research-tmp/skills/papers-research ~/.claude/skills/
rm -rf ~/.claude/skills/papers-research-tmp

# Windows (PowerShell)
git clone https://github.com/xwmxcz/papers-skill $env:USERPROFILE\.claude\skills\papers-research-tmp
Copy-Item -Recurse $env:USERPROFILE\.claude\skills\papers-research-tmp\skills\papers-research $env:USERPROFILE\.claude\skills\
Remove-Item -Recurse -Force $env:USERPROFILE\.claude\skills\papers-research-tmp

You'll need to manually edit SKILL.md to replace ${CLAUDE_PLUGIN_ROOT} with the absolute path to the skill's scripts/ directory, since the variable is only expanded by the plugin loader.

Option 3 — Local development

git clone https://github.com/xwmxcz/papers-skill
claude --plugin-dir ./papers-skill

Single-session, no permanent install.

---

Dependencies

The bundled Python script needs three packages:

pip install httpx arxiv PyMuPDF

The skill will tell you if any are missing.

---

Project structure

papers-skill/
├── .claude-plugin/
│   ├── plugin.json           # plugin manifest
│   └── marketplace.json      # so the repo is its own marketplace
├── skills/
│   └── papers-research/
│       ├── SKILL.md          # workflow doc Claude reads
│       └── scripts/
│           └── papers.py     # standalone CLI (the actual logic)
├── README.md
├── CHANGELOG.md
└── LICENSE

The skill uses ${CLAUDE_PLUGIN_ROOT} to resolve papers.py at runtime, so it works regardless of where the plugin is installed on disk.

---

How it compares to papers-mcp

	papers-mcp	papers-skill
Format	MCP server	Claude Code plugin / Skill
Process	Long-running stdio process	One-shot Python call per command
Install	claude mcp add ...	/plugin install ...
Dependencies	fastmcp + others	Just httpx, arxiv, PyMuPDF
Sharing	Each user configures manually	One-line plugin install
Latency	µs (process resident)	~100-500ms cold start per call
Best for	Heavy / frequent use	Occasional research tasks

The two are interchangeable feature-wise and can be installed side by side.

---

Acknowledgements

Derived from papers-mcp by the same author, originally inspired by the broader academic-paper-MCP ecosystem.

Built on:

• Semantic Scholar Academic Graph API
• arXiv API via the arxiv Python client
• PyMuPDF for PDF text extraction

---

License

MIT — see LICENSE.