Vision OCR MCP Server

A Model Context Protocol server for extracting text from images. This server enables LLMs to read invoices, receipts, and documents in 100+ languages while preserving the original script.

About this MCP Server: To understand how to connect to and utilize this MCP server, please refer to the official Model Context Protocol documentation at mcp.apify.com.

---

Connection URL

MCP clients can connect to this server at:

https://accelerationengg--vision-ocr-mcp.apify.actor/mcp

---

Client Configuration

To connect to this MCP server, use the following configuration in your MCP client:

{
  "mcpServers": {
    "vision-ocr": {
      "url": "https://accelerationengg--vision-ocr-mcp.apify.actor/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APIFY_TOKEN"
      }
    }
  }
}

Note: Replace YOUR_APIFY_TOKEN with your actual Apify API token. You can find your token in the Apify Console.

---

Claude Desktop Configuration

To use this MCP server with Claude Desktop, add the following configuration to your Claude Desktop settings:

Location: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)

{
  "mcpServers": {
    "apifyVisionOCR": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://accelerationengg--vision-ocr-mcp.apify.actor/mcp",
        "--header",
        "Authorization: Bearer YOUR_APIFY_TOKEN"
      ]
    }
  }
}

Steps:

1. Open Claude Desktop configuration file at the location above
2. Add the configuration with your Apify API token (replace YOUR_APIFY_TOKEN)
3. Save the file
4. Restart Claude Desktop
5. The vision_ocr tool will now be available in your conversations

---

Available Tools

vision_ocr - Extracts structured data from images with language detection and price extraction.

Parameters:

• images (array, required) - List of image URLs, file paths, or base64 strings (max 15)
• output_format (string, optional) - "json" (default) or "toon" for compact output

Returns:

{
  "language_detected": "ur",
  "description_text": "رسید | تاریخ: ۲۰۲۶-۰۱-۰۴ | چائے",
  "price_1": "₨۵۰",
  "price_2": "₨۱۲۵"
}

---

Features

✅ Multilingual OCR - Urdu (اردو), Arabic (العربية), English, Chinese (中文), and 100+ languages
✅ Price Detection - Automatically extracts prices from invoices/receipts
✅ Layout Preservation - Maintains tables and columns with "|" separators
✅ Batch Processing - Process up to 15 images in parallel
✅ Fast - 12-15 seconds per image

---

Supported Formats

Images: PNG, JPG, JPEG, WEBP (GIF not supported)
Languages: 100+ including Urdu, Arabic, English, Chinese, Hindi, Spanish, French, German

Output Formats

The server supports two output formats optimized for different use cases:

JSON Format (Default)

Standard structured output - easiest to parse and integrate with applications.

Example Output:

{
  "model": "Qwen/Qwen3-VL-30B",
  "image_count": 1,
  "total_time_seconds": 3.91,
  "results": [
    {
      "index": 0,
      "data": {
        "language_detected": "ar",
        "description_text": "TURKISH CORNER Date:6/10/2019 Time:6:56 PM Table:B12 Ticket No:243 -1Homus حمص 1-Mutabel متبل 1-Baba Ghanouj بابا غنوج 1-Fatoush فتوش 1-Olive Salad سلطة زيتون 1-Green Salad سلطة خضراء 1-Grapes Leaves ورق عنب 1-Tabouleh تبولة 1-Vegetable with Youghurt Salad سلطة خضار باللبن 1-Hot Salad سلطة حارة Total: 8.00 Cash 8.00 THANK YOU",
        "price_1": "8.00",
        "price_2": "8.00"
      },
      "processing_time": 3.91
    }
  ]
}

TOON Format (Token-Efficient)

Compact notation that saves ~30% tokens - ideal for LLM processing and cost optimization.

Example Output:

model: Qwen/Qwen3-VL-30B-A3B-Instruct
image_count: 1
total_time_seconds: 3.76
results:
  [1]{index,data,processing_time}:
    0,{'language_detected': 'ar', 'description_text': 'TURKISH CORNER Date:6/10/2019 Time:6:56 PM Table:B12 Ticket No:243 -1Homus حمص 1-Mutabel متبل 1-Baba Ghanouj بابا غنوج 1-Fatoush فتوش 1-Olive Salad سلطة زيتون 1-Green Salad سلطة خضراء 1-Grapes Leaves ورق عنب 1-Tabouleh تبولة 1-Vegetable with Youghurt Salad سلطة خضار باللبن 1-Hot Salad سلطة حارة Total: 8.00 Cash 8.00 THANK YOU', 'price_1': '8.00', 'price_2': '8.00'},3.76

When to use each format:

• JSON: Standard API integration, automated parsing, strict schema validation
• TOON: Sending to LLMs for analysis, reducing token costs, human-readable logs

---

Use Cases

• Financial documents: Invoices, receipts, bills
• Multi-column tables: Spreadsheets, reports
• Multilingual documents: Documents with Arabic, Urdu, Chinese, and other scripts
• Form extraction: Structured data from forms

---

Python API Usage

Installation

$pip install apify-client

Basic Example

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")

# Extract text from image
run_input = {
    "images": ["https://example.com/receipt.jpg"],
    "output_format": "json"  # or "toon"
}

run = client.actor("accelerationengg/vision-ocr-mcp").call(run_input=run_input)

# Get results
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    text = item['results'][0]['data']['description_text']
    language = item['results'][0]['data']['language_detected']
    print(f"Language: {language}\nText: {text}")

Batch Processing

# Process multiple images
run_input = {
    "images": [
        "https://example.com/invoice1.jpg",
        "https://example.com/invoice2.jpg",
        "https://example.com/invoice3.jpg"
    ],
    "output_format": "json"
}

run = client.actor("accelerationengg/vision-ocr-mcp").call(run_input=run_input)

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    for result in item['results']:
        print(f"Image {result['index']}: {result['data']['description_text'][:100]}...")

TOON Format for LLM Processing

# Use TOON format to save ~30% tokens
run_input = {
    "images": ["https://example.com/receipt.jpg"],
    "output_format": "toon"
}

run = client.actor("accelerationengg/vision-ocr-mcp").call(run_input=run_input)

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    toon_output = item['content']
    
    # Send directly to Claude for analysis
    # Uses ~30% fewer tokens than JSON
    # response = claude.messages.create(
    #     model="claude-3-5-sonnet-20241022",
    #     messages=[{
    #         "role": "user",
    #         "content": f"Analyze this receipt:\n{toon_output}"
    #     }]
    # )

---

Example Usage

Single Image

Extract text from this receipt:
https://example.com/receipt.jpg

Multiple Images

Process these invoices:
- https://example.com/invoice1.jpg
- https://example.com/invoice2.jpg
- https://example.com/invoice3.jpg

---

Built with Qwen-VL, FastMCP, Apify