Invoice Extraction MCP Server

Extract structured data from invoice and receipt images using Vision Language Model. Dual-mode Actor:

• MCP Mode: Connect from Claude Desktop via Standby URL
• Direct Mode: Call via Apify API with imageUrl/imageUrls input

Setup & Usage

Mode 1: MCP Server (Claude Desktop)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "invoice-extraction": {
      "url": "https://ntriqpro--invoice-extraction-mcp.apify.actor/mcp"
    }
  }
}

Restart Claude Desktop. The 4 MCP tools will be available immediately.

Mode 2: Direct API (Apify Actor)

Call via Apify with single invoice:

curl -X POST "https://api.apify.com/v2/acts/[actor-id]/runs" \
  -H "Authorization: Bearer $APIFY_TOKEN" \
  -d '{"imageUrl": "https://example.com/invoice.jpg"}'

Or batch (up to 10 invoices):

curl -X POST "https://api.apify.com/v2/acts/[actor-id]/runs" \
  -H "Authorization: Bearer $APIFY_TOKEN" \
  -d '{"imageUrls": ["url1", "url2", ..., "url10"]}'

Results saved to Apify dataset.

4 Tools

Tool	Description	Price
extract_invoice	Extract vendor, date, items, tax, total from invoice image	$0.05
extract_receipt	Extract store, items, subtotal, tax, total, payment method from receipt	$0.05
classify_document	Classify document type (invoice, receipt, purchase order, credit note)	$0.05
batch_extract	Process multiple invoices at once (max 10)	$0.05 per image

How It Works

This is a real MCP server using:

• Apify Standby mode (always-on, no cold start)
• Streamable HTTP transport (/mcp endpoint)
• @modelcontextprotocol/sdk for protocol compliance
• Qwen2.5VL:7B vision model via local ntriq Vision API

Unlike regular Apify actors that require API calls, this server speaks native MCP protocol — Claude Desktop connects directly.

Vision API Integration

All image analysis powered by local Qwen2.5VL:7B vision model running at https://vision.ntriq.co.kr:

• No external AI APIs
• Instant response times
• Complete data privacy (images not stored after processing)
• Transparent pricing with per-event billing

Extraction Examples

Invoice Extraction

Input: Invoice image URL Output:

{
  "status": "success",
  "invoice": {
    "vendor": "ACME Corp",
    "vendorAddress": "123 Main St, Springfield, IL",
    "invoiceNumber": "INV-2024-001",
    "date": "2024-03-27",
    "dueDate": "2024-04-27",
    "lineItems": [
      {
        "description": "Consulting Services",
        "quantity": 40,
        "unitPrice": 150.00,
        "amount": 6000.00
      }
    ],
    "subtotal": 6000.00,
    "tax": 480.00,
    "taxRate": 8,
    "total": 6480.00,
    "currency": "USD",
    "paymentTerms": "Net 30"
  }
}

Receipt Extraction

Input: Receipt image URL Output:

{
  "status": "success",
  "receipt": {
    "storeName": "Coffee Shop ABC",
    "storeLocation": "Downtown",
    "receiptNumber": "RCP-0123456",
    "date": "2024-03-27",
    "time": "14:35",
    "lineItems": [
      { "description": "Espresso", "quantity": 1, "unitPrice": 4.50, "amount": 4.50 },
      { "description": "Croissant", "quantity": 2, "unitPrice": 3.00, "amount": 6.00 }
    ],
    "subtotal": 10.50,
    "tax": 0.84,
    "total": 11.34,
    "currency": "USD",
    "paymentMethod": "credit_card"
  }
}

Document Classification

Input: Document image URL Output:

{
  "status": "success",
  "classification": {
    "document_type": "invoice",
    "confidence": 0.98,
    "reason": "Contains invoice number, line items, tax calculation, and payment terms",
    "features_detected": ["invoice_number", "line_items", "tax_calculation", "payment_terms", "vendor_info"]
  }
}

Batch Processing

Process up to 10 invoices in a single call:

{
  "imageUrls": [
    "https://example.com/invoice1.jpg",
    "https://example.com/invoice2.jpg",
    "https://example.com/invoice3.jpg"
  ]
}

Response includes success/fail status per image and accumulated charges.

Pricing

Pay-Per-Event (PPE) billing — only charged when extraction succeeds:

Tool	Price	Notes
extract_invoice	$0.05	Per invoice extraction
extract_receipt	$0.05	Per receipt extraction
classify_document	$0.05	Per document classification
batch_extract	$0.05	Per image in batch (max 10 images)

No charge on errors. Success is required to trigger billing.

Legal Notice

Data Processing & Privacy (GDPR)

• Invoice images are processed on our AI server located in South Korea.
• South Korea has received an EU adequacy decision for data protection.
• Extracted data is returned to your Apify dataset only. We do not retain your invoice data.
• All uploaded images are automatically deleted within 24 hours of processing.
• We do not sell, share, or use your data for any purpose beyond the requested extraction.

Your Responsibilities (IMPORTANT)

• Before uploading, redact or mask any sensitive information you do not want processed:
  • Credit card numbers, bank account details
  • Social security numbers or national ID numbers
  • Any data not relevant to invoice extraction
• You are responsible for obtaining necessary consent from data subjects whose personal information appears on invoices.
• For EU/EEA users: By using this service, you consent to data processing as described above. You may request data deletion at any time.

Accuracy Disclaimer

• Extraction accuracy is approximately 75% for structured fields (amounts, dates, vendor names).
• AI-generated results must be verified by a human before use in accounting, tax, or legal contexts.
• This tool is an extraction assistant, not a certified accounting or tax preparation system.

Financial Data

• This Actor does not process payment card data (credit card numbers, CVV, expiration dates).
• Invoice amounts, vendor names, and dates are not subject to PCI DSS requirements.
• Users are responsible for compliance with their local tax and financial regulations.

Technical Details

MCP Protocol

This server implements the full Model Context Protocol specification:

• Stateless HTTP transport
• Request/response JSON-RPC 2.0
• Schema validation with Zod
• Error handling and timeout protection

Vision API

• Model: Qwen2.5VL:7B
• Max tokens: 2000 per extraction
• Timeout: 120 seconds default
• Format: JSON extraction with fallback text parsing

Supported Image Formats

• JPEG
• PNG
• WEBP
• GIF (single frame)
• PDF (single page when converted to image)

Maximum image size: 20MB

Use Cases

• Expense Management: Automate receipt processing for accounting
• Invoice Processing: Extract vendor and payment details for AP automation
• Document Classification: Route documents to appropriate workflows
• Data Entry Automation: Populate forms from physical documents
• Audit Trail: Build compliance-ready records from receipts

Requirements

• Claude Desktop (2.0+) or compatible MCP client
• Internet connection (to reach https://vision.ntriq.co.kr and image URLs)
• HTTPS URLs for all images (mixed content blocked)

Error Handling

Common error scenarios:

Error	Cause	Solution
Vision API error: 401	Auth failure	Check NTRIQ_VISION_URL environment variable
Vision API error: 404	Invalid image URL	Verify image URL is publicly accessible
Maximum 10 images per batch	Batch size too large	Split into multiple requests
imageUrls must be a non-empty array	Empty batch	Provide at least one image URL
Timeout	Processing took too long	Retry or use smaller image

Roadmap

• OCR confidence scores per field
• Multi-language extraction
• Custom field extraction prompts
• Handwriting detection
• Table detection and extraction
• Purchase order processing
• Expense categorization

Support

For issues, feature requests, or questions:

• GitHub: ntriq-public
• Email: support@ntriq.co.kr
• Telegram: @ntriq_support

Legal & Compliance

Data Processing Agreement

Available for enterprise customers. Contact sales@ntriq.co.kr

Privacy Policy

See privacy.ntriq.co.kr

Terms of Service

See terms.ntriq.co.kr

License

Proprietary — ntriq Inc. All rights reserved.