Book Capture

Capture books from Kindle, Apple Books, or PDF — then OCR and generate structured Markdown.

A Claude Code plugin that captures book pages as screenshots, extracts text via OCR, and generates thematically organized Markdown documents. Built for Obsidian knowledge vaults but works with any Markdown-based system.

What It Does

1. Captures every page of a book as screenshots (Mac Kindle, Apple Books, Kindle Cloud Reader, or PDF)
2. Extracts text via macOS Vision OCR + Claude Code agents for low-confidence pages
3. Generates 8-14 thematically organized Markdown files with rich formatting (tables, blockquotes, cross-references)
4. Creates a hub file with frontmatter and wikilinks to all topic files

The entire pipeline runs locally with no external API keys — OCR and content generation use Claude Code's built-in capabilities.

Supported Platforms

Installation

Install from GitHub (recommended)

In Claude Code, run:

/plugins

1. Navigate to the Marketplaces tab
2. Select Add marketplace and enter: masterleopold/book-capture
3. Navigate to the Discover tab
4. Find book-capture and install it

After installation, the commands are available in every Claude Code session.

Try It (one-time)

git clone https://github.com/masterleopold/book-capture.git
claude --plugin-dir ./book-capture

First-Time Setup

After installing, run the setup script to install Node.js dependencies and compile the Vision OCR binary:

bash ~/.claude/plugins/cache/book-capture-marketplace/book-capture/*/scripts/setup.sh

Or the plugin will auto-detect and prompt you on first use.

Requirements

• macOS (required for screencapture and Vision OCR)
• Claude Code CLI installed and authenticated
• Node.js 20+
• Xcode Command Line Tools (xcode-select --install)
• Accessibility permission for Terminal/Claude Code (System Settings > Privacy & Security > Accessibility)
• Poppler (PDF only): brew install poppler

Commands

Source-Specific (recommended)

Pipeline Steps

Quick Start

/book-capture:kindle B0883TQ3ZN

Claude Code will:

1. Ask for book title, author, category, and location
2. Remind you to open the book in Kindle to the first page
3. Capture all pages (auto-stops at end of book)
4. Run Vision OCR + agent re-reading for low-confidence pages
5. Analyze content and create 8-14 thematic topic files
6. Generate a hub file with wikilinks

How It Works

Screenshot Capture

Each platform uses macOS-native tools:

• CGWindowList (via inline Swift) to find the app window ID
• screencapture to capture individual window frames
• AppleScript to control page navigation (Page Down for Kindle, arrow keys for Books)
• Duplicate detection to auto-stop at end of book (3 consecutive identical pages)

OCR Pipeline

1. macOS Vision OCR (fast, local) processes all pages with confidence scoring
2. Pages below the confidence threshold are re-read by Claude Code agents using multimodal image reading
3. Results are merged into raw_text.json

Note: macOS Vision cannot read vertical Japanese text (tategaki). For vertical text books, all pages are re-read by agents.

Markdown Generation

1. Claude Code analyzes the full text and identifies 8-14 thematic categories (organized by information type, not original chapter order)
2. Parallel agents generate detailed topic files (300-600 lines each) with:
   • Genre-specific structure (business, technical, humanities, science, narrative)
   • Tables, blockquotes, bold key terms, cross-references
   • [[wikilinks]] between sibling topics
3. A hub file is created with frontmatter and links to all topics

Output Structure

Books/entries/BookTitle.md              # Hub file with frontmatter
Knowledge/Category/BookTitle/
  01_Theme_Name.md                      # Topic file (300-600 lines)
  02_Theme_Name.md
  ...
  10_Theme_Name.md

Hub File Example

---
tags:
  - source/book
  - type/framework
  - theme/fundraising
Category: "Startup"
Rating: ""
author: "Author Name"
Location: "Knowledge/Startup/BookTitle"
Chapters: 10
Language: "EN"
URL: "https://www.amazon.co.jp/dp/B0883TQ3ZN"
---

# Book Title

Book summary (2-3 sentences).

## Topics

- [[01_Theme Name]] - Description
- [[02_Theme Name]] - Description
...

Configuration

Per-project settings via .claude/book-capture.local.md:

---
vault_root: /path/to/obsidian/vault
captures_dir: Books/files/book-captures
entries_dir: Books/entries
default_source: kindle
default_language: JP
default_dpi: 200
---

If no settings file exists, the plugin auto-detects Obsidian vaults by looking for .obsidian/ in the current directory or parents.

Troubleshooting

Architecture

book-capture/
  .claude-plugin/
    plugin.json           # Plugin manifest
    marketplace.json      # Marketplace metadata
  commands/               # 7 slash commands
    kindle.md             # /book-capture:kindle
    books.md              # /book-capture:books
    cloud.md              # /book-capture:cloud
    pdf.md                # /book-capture:pdf
    capture.md            # /book-capture:capture (generic)
    ocr.md                # /book-capture:ocr
    generate.md           # /book-capture:generate
  agents/
    ocr-reader.md         # Multimodal OCR re-reader
    content-writer.md     # Thematic content generator
  skills/
    book-capture/
      SKILL.md            # Auto-activating skill
  scripts/
    capture-kindle-mac.mjs
    capture-books-app.mjs
    kindle-capture.mjs    # Playwright Cloud Reader
    capture-pdf.mjs
    extract-text.mjs      # Vision OCR
    generate-markdown.mjs # Direct API fallback
    book-capture-utils.mjs
    vision-ocr.swift      # macOS Vision CLI
    setup.sh              # Dependency installer
    package.json
  templates/
    settings-template.md

License

MIT