---
name: document-tools
description: "Document processing: PDF editing (nano-pdf) and text extraction from PDFs/scans (pymupdf, marker-pdf)."
version: 1.0.0
author: Hermes Agent
license: MIT
metadata:
  hermes:
    tags: [pdf, document, ocr, text-extraction, nano-pdf, pymupdf, marker]
    related_skills: [powerpoint, google-workspace]
---

# Document Tools

Two document processing workflows: in-place PDF editing and text extraction from PDFs/scans.

## Section 1: nano-pdf — In-Place PDF Editing

Edit PDF text, fix typos, and update titles using the `nano-pdf` CLI with natural language prompts.

**When to use:** Fixing typos in PDFs, updating document titles, changing specific text in PDFs without regenerating.

**Installation:**
```bash
npm install -g nano-pdf
```

**Usage:**
```bash
# Fix a typo
nano-pdf document.pdf "change 'teh' to 'the'"

# Update title
nano-pdf document.pdf "set title to 'Annual Report 2025'"
```

**See:** `references/nano-pdf.md` for full CLI reference.

## Section 2: OCR & Document Text Extraction

Extract text from PDFs and scanned documents using pymupdf (fast native extraction) or marker-pdf (OCR + layout analysis).

**When to use:** Reading PDF content, extracting text from scans, converting PDFs to plain text or markdown.

### pymupdf (Fast, Native PDF)

```python
import pymupdf  # fitz
doc = pymupdf.open("document.pdf")
for page in doc:
    text = page.get_text()
    print(text)
```

**Best for:** Text-based PDFs, fast extraction, page-level access, image/MRI extraction.

### marker-pdf (OCR + Layout)

```bash
pip install marker-pdf
marker_single document.pdf output_dir
```

**Best for:** Scanned documents, OCR-needed PDFs, preserving layout, converting to markdown.

### Choosing

| Need | Tool |
|------|------|
| Fast text extraction from text PDF | pymupdf |
| Scanned/image PDFs | marker-pdf |
| Need markdown output | marker-pdf |
| Extract images from PDF | pymupdf |
| Preserve document layout | marker-pdf |

**See:** `references/ocr-and-documents.md` for full API reference and troubleshooting.