# pdf-chapter-splitter

An MCP (Model Context Protocol) server that splits PDFs by chapters/sections and reads them in Claude-friendly chunks. Designed for use with Claude Code to enable structured reading of large PDF documents.

PDFをチャプター単位で分割し、Claudeが扱いやすいサイズで読み込めるMCPサーバーです。Claude Codeと組み合わせて、大きなPDFの構造化された読み込みを実現します。

## How it works / 仕組み

```
[You / ユーザー] → provide a PDF file path / PDFファイルパスを指定
  ↓
[pdf-chapter-splitter MCP] → analyze structure & extract text / 構造解析＆テキスト抽出
  ↓
[Claude Code] ← receives chapter text + metadata / チャプターテキスト+メタデータで返却 → analyzes content / 内容を分析
```

## Prerequisites / 前提条件

- **Node.js** >= 18

No external binaries required. Uses `pdfjs-dist` (pure JavaScript).

外部バイナリは不要です。`pdfjs-dist`（ピュアJavaScript）で動作します。

## Installation / インストール

### Claude Code CLI (Recommended / 推奨)

```bash
claude mcp add pdf-chapter-splitter -- npx -y pdf-chapter-splitter
```

This registers the MCP server with Claude Code. No manual configuration needed.

### npx (Direct execution / 直接実行)

```bash
npx -y pdf-chapter-splitter
```

### Manual installation / 手動インストール

```bash
git clone https://github.com/keigoly/pdf-chapter-splitter.git
cd pdf-chapter-splitter
npm install
npm run build
```

If you installed manually, add the following to your `.mcp.json` (project-level) or `~/.claude/settings.json` (global):

```json
{
  "mcpServers": {
    "pdf-chapter-splitter": {
      "command": "node",
      "args": ["/path/to/pdf-chapter-splitter/dist/index.js"]
    }
  }
}
```

## Tools / ツール

### `get_pdf_info`

Get PDF metadata (title, author, page count, file size).

| Parameter | Type | Description |
|---|---|---|
| `file_path` | string (required) | Absolute path to the PDF file |

### `get_toc`

Extract table of contents (bookmarks/outline) from a PDF. Returns hierarchical chapter structure with page numbers.

| Parameter | Type | Description |
|---|---|---|
| `file_path` | string (required) | Absolute path to the PDF file |

### `list_chapters`

List chapters/sections detected from the PDF. Uses TOC if available, otherwise splits by pages.

| Parameter | Type | Default | Description |
|---|---|---|---|
| `file_path` | string (required) | - | Absolute path to the PDF file |
| `level` | number (optional) | top-level only | Max TOC depth level to include (0=top level only) |

### `read_chapter`

Read the text content of a specific chapter by its index (from `list_chapters`).

| Parameter | Type | Description |
|---|---|---|
| `file_path` | string (required) | Absolute path to the PDF file |
| `chapter_index` | number (required) | Chapter index from `list_chapters` output |
| `level` | number (optional) | Max TOC depth level (must match `list_chapters`) |

### `read_pages`

Read text content from a specific page range.

| Parameter | Type | Description |
|---|---|---|
| `file_path` | string (required) | Absolute path to the PDF file |
| `start_page` | number (required) | Start page number (1-based) |
| `end_page` | number (required) | End page number (1-based, inclusive) |

### `detect_headings`

Detect heading candidates by analyzing font sizes. Useful for PDFs without TOC/bookmarks.

| Parameter | Type | Description |
|---|---|---|
| `file_path` | string (required) | Absolute path to the PDF file |

## 3-level fallback strategy / 3段階フォールバック

1. **TOC/Bookmarks** → Use as chapter structure / そのままチャプター構造として利用
2. **Font size analysis** → Detect headings from larger text / フォントサイズからヘディング検出
3. **Auto-split** → Split every 20 pages / 20ページ単位で自動分割

## Constraints / 制約

| Item | Limit |
|------|-------|
| Supported format | `.pdf` |
| Max response size | 200,000 chars / 1,800 lines |
| Auto-split chunk | 20 pages |
| External deps | None (pdfjs-dist only) |

## Use cases / ユースケース

- **Structured reading**: Read technical books chapter by chapter / 技術書をチャプターごとに読む
- **Document review**: Review specific sections of specs or reports / 仕様書やレポートの特定セクションをレビュー
- **Paper analysis**: Analyze academic papers section by section / 学術論文をセクション単位で分析
- **Cross-PDF search**: Get TOCs from multiple PDFs to find relevant chapters / 複数PDFの目次から関連章を検索

## License

MIT