The Hidden Risk in AI-Assisted Work
Every time you use ChatGPT, Claude, or Cursor, you might be sharing sensitive information. Most users don't realize their data is being sent to third-party AI providers.
- Customer names and emails
- Financial account numbers
- Medical records and patient IDs
- Trade secrets and proprietary data
How It Works: Automatic Protection
You Write Normally
With sensitive data
AI Tool
Cursor, Claude, etc.
AI Processes
Safe placeholders only
Restore Values
Original data back
The MCP Server acts as a privacy shield between your AI tools and sensitive data. AI never sees your real information.
What Gets Protected?
The system automatically detects and protects 320+ types of sensitive information across 70+ countries
People
John Doe, Jane Smith, Dr. Williams
Emails
john@example.com, info@company.org
Phones
+1-555-1234, (555) 123-4567
Credit Cards
4532-1234-5678-9010
Addresses
123 Main Street, New York, NY
Dates
January 15, 2026, 01/15/2026
Medical Info
Patient ID, prescription numbers
Financial IDs
IBANs, Tax IDs, VAT numbers
Digital Identifiers
IP addresses, MAC addresses, URLs
And 280+ More Types
License plates, passport numbers, etc.
Key Benefits
10 Privacy Tools
Complete toolkit: analyze text, anonymize, detokenize, batch analyze (1-100 texts), image analysis, image redaction, balance check, cost estimation, and session management.
Batch Processing
Analyze 1-100 texts in a single request with parallel server-side processing. Each item can have its own language and entity settings.
Image OCR Redaction
Detect and redact PII in images via Tesseract OCR (38 languages). Send base64-encoded images, receive redacted PNG with black/white/colored fills.
48 Languages Supported
Detects PII in English, German, French, Spanish, Chinese, Japanese, Arabic, and 41 other languages. RTL support included.
Personal Encryption Keys
Load your own AES-128/192/256 keys for encrypt operator. Decrypt anytime with your key. Keys stored securely in your account.
Native Claude Desktop + HTTP
Stdio transport for Claude Desktop with zero network latency. HTTP endpoint for Cursor, VS Code, and any MCP-compatible tool.
Setup in 2 Minutes
Choose stdio for Claude Desktop or HTTP for other tools
Claude Desktop (Recommended):
- Install Node.js 18+ if not already installed
- Run: npx cloak-business-mcp-server
- Add to claude_desktop_config.json with your API key
- Restart Claude Desktop - 10 privacy tools available!
HTTP (Cursor, VS Code, etc.):
- Get your API key from Settings → API Access
- Add MCP server URL: https://cloak.business/mcp
- Configure Authorization header with Bearer token
- No installation required - works instantly
MCP Server vs Manual Redaction
| Aspect | Manual Redaction | cloak.business MCP |
|---|---|---|
| Speed | Minutes per document | < 1 second |
| Accuracy | Human error-prone | 95%+ AI accuracy |
| Reversible | No - permanent | Yes - fully reversible |
| Maintenance | Constant vigilance | Automatic |
| Languages | One at a time | 48 simultaneously |
| Integration | Copy/paste workflow | Seamless in your tools |
| Cost | Staff time | From €0/month |
Is This Right for You?
Best For
- ✦Dev teams using Claude Code or Cursor AI that must prevent PII from entering AI coding assistants
- ✦CI/CD pipelines where code reviews or PR automation touch data containing customer PII
- ✦Organizations requiring audit logs of every AI prompt containing personal data for GDPR compliance
- ✦Multi-model AI workflows (Claude, GPT-4, custom LLMs) needing consistent PII removal at the IDE layer
Not Ideal For
- ✦Browser-based AI tools (ChatGPT web, Claude.ai web) — use the Chrome Extension instead
- ✦Batch processing of historical document archives — use the Web App batch tool instead
- ✦Teams not yet using AI coding assistants in their development workflow
Frequently Asked Questions
Which integration should I use - stdio or HTTP?
For Claude Desktop, use stdio (npx command) for best performance. For Cursor, VS Code extensions, or custom apps, use HTTP (just add URL). Both support all 10 tools.
What tools are available?
10 tools: analyze_text, anonymize_text, detokenize_text, batch_analyze (1-100 texts), analyze_image, redact_image, get_balance, estimate_cost, list_sessions, delete_session.
How does batch processing work?
Use batch_analyze to process 1-100 texts in a single request. Each item can have its own language and entity settings. Results include per-item entities plus aggregated batch metadata.
Can MCP anonymize images?
Yes! Use analyze_image to detect PII with bounding boxes, or redact_image to remove PII. Send base64-encoded PNG/JPEG/BMP/TIFF (max 10MB). Redacted images return as base64 PNG with configurable fill colors.
What's the difference between redact and tokenize modes?
Redact mode permanently removes PII. Tokenize mode replaces with reversible tokens and returns a session_id. Use detokenize_text with the session_id to restore original values.
How do I test my MCP connection?
Ask your AI assistant: 'What's my cloak.business token balance?' If it returns your balance, the connection is working. You can also use estimate_cost to check processing costs before analyzing.