PrivaScan
Privacy-first document intelligence for Claude. OCR, barcode reading, document scanning, image classification, and face detection — all processed on-device using Apple Vision. Nothing ever leaves your Mac.
Why PrivaScan?
Most document processing tools send your files to the cloud. PrivaScan runs entirely on your machine using Apple's Vision framework. Your medical bills, contracts, tax forms, and personal documents stay private.
- On-device OCR — Text recognition powered by Apple Vision, no network calls
- Document detection — Automatically finds and crops documents in photos
- PDF extraction — Page-by-page text extraction with language support
- Barcode/QR reading — Decode any barcode type from images
- Image classification — Identify document types before processing
- Face detection — Find faces for PII redaction (detection only, no recognition)
Requirements
- macOS 14.0+
- Swift 6.0+ / Xcode 16+
Install
Claude Marketplace
claude install alpozcan/PrivaScan
Build from source
git clone https://github.com/alpozcan/PrivaScan.git
cd PrivaScan
swift build -c release
The binary will be at .build/release/PrivaScan.
Setup
Claude Code
Add to your MCP config (~/.claude.json):
{
"mcpServers": {
"privascan": {
"command": "/path/to/PrivaScan/.build/release/PrivaScan"
}
}
}
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"privascan": {
"command": "/path/to/PrivaScan/.build/release/PrivaScan"
}
}
}
Tools
recognize_text
Extract text from an image using on-device OCR.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to image file |
languages | string[] | no | Recognition languages (default: ["en-US"]) |
Supported formats: PNG, JPEG, TIFF, HEIC
scan_document
Detect document regions in a photo, crop, and OCR.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to image file |
languages | string[] | no | Recognition languages |
extract_pdf_text
Extract text from PDF pages using on-device OCR.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to PDF file |
pages | string | no | Page range (e.g. "1-5", "1,3,7") |
languages | string[] | no | Recognition languages |
read_barcodes
Detect and decode barcodes and QR codes.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to image file |
Supported types: QR, EAN-13, EAN-8, Code 128, Code 39, UPC-E, PDF417, Aztec, Data Matrix
classify_image
Classify an image using Apple's built-in Vision model.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to image file |
max_results | integer | no | Max results to return (default: 5) |
detect_faces
Detect faces and return bounding boxes. Useful for identifying PII in documents like IDs or passports. Detection only — no face recognition.
| Parameter | Type | Required | Description |
|---|
file_path | string | yes | Absolute path to image file |
Try It
The demo/ folder includes realistic sample documents (medical form, W-2, employment contract, bank statement) for a privacy audit scenario. After setup, try:
Scan all files in demo/ for personal information. Cross-reference what you find across documents and give me a privacy risk report.
See demo/DEMO.md for more prompts.
Privacy
PrivaScan processes everything locally using Apple's Vision framework. No data is sent to any server for OCR, classification, or detection. The only data that leaves your machine is the extracted text that gets passed back to Claude through the MCP protocol — never the original images or documents.
License
MIT
alpozcan