mcpbeat

PDF OCR Extraction

claude-office-skills/pdf ocr extraction

Extract text from scanned PDFs using optical character recognition

2k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
353
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/claude-office-skills/skills --skill PDF OCR Extraction

The instruction itself

28 sections, as written by the author

PDF OCR Extraction

Extract text from scanned documents and image-based PDFs using OCR technology.

Overview

This skill helps you:

  • Extract text from scanned documents
  • Make image PDFs searchable
  • Digitize paper documents
  • Process handwritten text (limited)
  • Batch process multiple documents

How to Use

Basic OCR

"Extract text from this scanned PDF"
"OCR this document image"
"Make this PDF searchable"

With Options

"Extract text from pages 1-10, English language"
"OCR this document, preserve layout"
"Extract and output as structured data"

Document Types

OCR Quality by Document Type

| Document Type | Expected Quality | Tips |

|---------------|------------------|------|

| Typed documents | ⭐⭐⭐⭐⭐ 95%+ | Best results |

| Printed books | ⭐⭐⭐⭐ 90%+ | Watch for aging |

| Forms | ⭐⭐⭐⭐ 85%+ | Check boxes may need manual |

| Tables/Data | ⭐⭐⭐ 80%+ | Structure may need fixing |

| Handwritten (neat) | ⭐⭐ 60-80% | Variable results |

| Handwritten (cursive) | ⭐ 30-60% | Often needs manual review |

| Mixed content | ⭐⭐⭐ 75%+ | Depends on complexity |

Output Formats

Plain Text Extraction

## OCR Result: [Document Name]

**Pages Processed**: [X]
**Language**: [Detected/Specified]
**Confidence**: [X]%

---

[Extracted text content here]

---

### Notes
- [Any issues or uncertainties]
- [Characters that may be incorrect]

Structured Extraction

## OCR Extraction: [Document Name]

### Document Info
| Field | Value |
|-------|-------|
| Title | [Extracted or inferred] |
| Date | [If found] |
| Author | [If found] |

### Content by Section

#### [Header 1]
[Content under this header]

#### [Header 2]
[Content under this header]

### Tables Found
| Column 1 | Column 2 | Column 3 |
|----------|----------|----------|
| [Data] | [Data] | [Data] |

### Uncertain Text
| Page | Original | Confidence | Possible |
|------|----------|------------|----------|
| 3 | "teh" | 70% | "the" |
| 5 | "l0ve" | 65% | "love" |

Searchable PDF Output

## OCR to Searchable PDF

**Source**: [filename.pdf]
**Output**: [filename_searchable.pdf]

### Processing Summary
| Metric | Value |
|--------|-------|
| Pages | [X] |
| Words extracted | [Y] |
| Average confidence | [Z]% |
| Processing time | [T] seconds |

### Quality Report
- [X] pages with 95%+ confidence
- [Y] pages with 80-94% confidence
- [Z] pages with <80% confidence (review recommended)

### Searchability
✅ Document is now text-searchable
✅ Original images preserved
✅ Text layer added behind images

Pre-Processing Tips

Image Quality Checklist

Before OCR, ensure:

  • [ ] Resolution: 300 DPI minimum (600 for small text)
  • [ ] Contrast: Clear black text on white background
  • [ ] Alignment: Document is straight (not skewed)
  • [ ] Completeness: No cut-off edges
  • [ ] Cleanliness: No stains, marks, or shadows

Common Pre-Processing Steps

| Issue | Solution |

|-------|----------|

| Low resolution | Upscale image first |

| Skewed/rotated | Auto-deskew |

| Poor contrast | Adjust levels/threshold |

| Noise/specks | Apply noise reduction |

| Shadows | Flatten lighting |

| Color document | Convert to grayscale |

Language Support

Supported Languages

  • Excellent: English, Spanish, French, German, Italian
  • Good: Chinese (Simplified/Traditional), Japanese, Korean
  • Moderate: Arabic, Hebrew (RTL support), Hindi
  • Basic: Many others with varying quality

Multi-Language Documents

"OCR this document, detect language automatically"
"Extract text, primary: English, secondary: Chinese"

Handling Specific Content

Forms and Checkboxes

## Form Extraction: [Form Name]

### Field Values
| Field | Value | Confidence |
|-------|-------|------------|
| Name | John Smith | 98% |
| Date | 01/15/2026 | 95% |
| Address | 123 Main St | 92% |

### Checkboxes
| Question | Checked |
|----------|---------|
| Option A | ☑️ Yes |
| Option B | ☐ No |
| Option C | ☑️ Yes |

### Signature
[Signature detected on page X - cannot extract text]

Tables

## Table Extraction

### Table 1 (Page 2)
| Header A | Header B | Header C |
|----------|----------|----------|
| Value 1 | Value 2 | Value 3 |
| Value 4 | Value 5 | Value 6 |

**Table confidence**: 85%
**Note**: Column 3 may have alignment issues

Handwritten Text

## Handwritten Text Extraction

**Legibility Assessment**: [Good/Fair/Poor]
**Recommended**: Manual review

### Extracted Text (Confidence: 65%)
[Extracted text with uncertain words marked]

### Uncertain Words
| Original | Best Guess | Alternatives |
|----------|------------|--------------|
| [image] | "meeting" | "meeting", "meaning" |
| [image] | "Tuesday" | "Tuesday", "Thursday" |

⚠️ **Low confidence extraction - please verify manually**

Batch Processing

Batch OCR Job

## Batch OCR Processing

**Folder**: [Path]
**Total Documents**: [X]
**Status**: [In Progress/Complete]

### Results
| File | Pages | Confidence | Status |
|------|-------|------------|--------|
| doc1.pdf | 5 | 96% | ✅ Complete |
| doc2.pdf | 12 | 88% | ✅ Complete |
| doc3.pdf | 3 | 72% | ⚠️ Review |
| doc4.pdf | 8 | - | ❌ Failed |

### Issues
- doc3.pdf: Pages 2-3 have handwriting
- doc4.pdf: File corrupted

### Summary
- Successful: [X]
- Need Review: [Y]
- Failed: [Z]

Tool Recommendations

Cloud Services

  • Google Cloud Vision (excellent accuracy)
  • Amazon Textract (good for forms)
  • Azure Computer Vision (balanced)
  • Adobe Acrobat (integrated)

Desktop Software

  • ABBYY FineReader (best accuracy)
  • Adobe Acrobat Pro (reliable)
  • Readiris (good value)
  • Tesseract (free, open source)

Programming Libraries

  • pytesseract (Python + Tesseract)
  • EasyOCR (Python, multi-language)
  • PaddleOCR (Python, good for Asian languages)

Limitations

  • Cannot guarantee 100% accuracy
  • Handwritten text has low accuracy
  • Very small text may not extract well
  • Decorative fonts are problematic
  • Background images reduce quality
  • Cannot read text in complex graphics
  • Processing time increases with pages

How to use it

Copy the folder

Take claude-office-skills/pdf ocr extraction from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.