PaperShelf

A Claude Code skill that turns your documents into an AI-ready knowledge base — in one command.

Drop in a PDF, a slide deck, a Word doc, or an image. PaperShelf converts it, writes a summary, and adds it to a searchable index. From then on, ask Claude anything about your collection — it navigates from the index to the right paper to the right paragraph, without loading everything at once.

---

How it works

You: "Add these papers to my knowledge base"
  ↓
PaperShelf converts each document to Markdown (via MinerU)
  ↓
Summarizes it (500–1000 words, adapted to the document type)
  ↓
Extracts an exact table of contents with line numbers
  ↓
Adds a reading note to your INDEX.md explaining why it matters
  ↓
You: "What does paper X say about topic Y?"
Claude: reads INDEX → summary → digs into full text if needed

---

Features

• One command to add — just tell Claude the file path, it handles everything
• Parallel processing — multiple documents are converted and summarized simultaneously
• Flexible summaries — adapts to what the document actually is (paper, report, manual, proposal…)
• Exact table of contents — section headings copied character-for-character with line numbers, so searches always find the right passage
• Project-aware index — each entry gets a note explaining why it's relevant to your project, not just what it says
• Supports most document types — PDF, Word, PowerPoint, images, HTML

---

Requirements

• Claude Code
• A MinerU API key — free tier available

---

Installation

Copy the skill into your Claude skills directory:

cp -r paper-shelf ~/.claude/skills/

Set your MinerU API key (either way works):

# As an environment variable
export MINERU_API_KEY=your_key_here

# Or in a .env file in your project root
echo 'MINERU_API_KEY=your_key_here' >> .env

---

Usage

Open Claude Code in your project directory and just talk to it:

Add papers/my_paper.pdf to the knowledge base

Add all the PDFs in my literature folder to the knowledge base

What does the UniCAS paper say about multi-scale feature extraction?

Based on my collected papers, help me brainstorm ideas for Chapter 2

---

Knowledge base layout

knowledge-base/
  INDEX.md                        ← Start here — categorized entries with reading notes
  papers/
    <Paper Title>/
      full.md                     ← Full converted text
      summary.md                  ← Summary + table of contents with line numbers
      images/                     ← Figures extracted from the document

---

Supported formats

.pdf .doc .docx .ppt .pptx .png .jpg .jpeg .html

---

License

MIT