OCR MCP Service

中文 | 日本語

MCP-based OCR service powered by PaddleOCR, supporting text recognition and table recognition.

Features

• OCR Text Recognition: Recognize text content in images, supporting Chinese, English, and other languages

• Table Recognition: Recognize tables in images and return structured data

• Handwriting Recognition: Recognize handwritten text in images

• Formula Recognition: Recognize mathematical formulas in images and return LaTeX format

• Multi-format Support: Support JPG, PNG, BMP, TIFF, WebP and other image formats

• URL Support: Support both local file paths and network URLs

• MCP Protocol: Standard MCP service, can be integrated into MCP-compatible AI assistants

Related MCP server: mcp_ocr

Installation

Install via pip

pip install ocr-mcp

Install from source

git clone https://github.com/LinuxLinking/OCR_MCP.git
cd OCR_MCP
pip install -e .

Usage

Run as MCP service

# Run directly
ocr-mcp

# Or run via Python module
python -m ocr_mcp

Integrate with Claude Desktop

Add to Claude Desktop configuration file:

{
  "mcpServers": {
    "ocr": {
      "command": "ocr-mcp",
      "args": []
    }
  }
}

MCP Tool Description

ocr_recognize

Recognize text content in images.

Parameters:

• source (required): Image source, local file path or URL

• language (optional): Recognition language, default ch (Chinese), supports en, japan, korean, etc.

• use_angle_cls (optional): Whether to enable text direction classification, default true

Example output:

1. Hello World (confidence: 98.50%)
2. 你好世界 (confidence: 97.20%)

ocr_table

Recognize table content in images and return structured data.

Parameters:

• source (required): Image source, local file path or URL

• language (optional): Recognition language, default ch

Example output:

+--------+-----+------+
| Name   | Age | City |
+--------+-----+------+
| Alice  | 25  | NYC  |
+--------+-----+------+
| Bob    | 30  | LA   |
+--------+-----+------+

ocr_handwrite

Recognize handwritten text content in images.

Parameters:

• source (required): Image source, local file path or URL

• language (optional): Recognition language, default ch

Example output:

1. Hello World (confidence: 85.50%)
2. 你好 (confidence: 82.30%)

ocr_formula

Recognize mathematical formulas in images and return LaTeX format.

Parameters:

• source (required): Image source, local file path or URL

Example output:

Formula 1:
  LaTeX: E = mc^2
Formula 2:
  LaTeX: \int_{0}^{\infty} e^{-x^2} dx = \frac{\sqrt{\pi}}{2}

Development

Install development dependencies

pip install -e ".[dev]"

Run tests

pytest tests/ -v

Project Structure

OCR_MCP/
├── src/ocr_mcp/
│   ├── __init__.py          # Package initialization
│   ├── __main__.py          # Module entry point
│   ├── server.py            # MCP service main entry (dispatch layer)
│   ├── ocr_engine.py        # Subprogram 1: Text recognition engine
│   ├── table_parser.py      # Subprogram 2: Table recognition engine
│   ├── formula_engine.py    # Subprogram 3: Formula recognition engine
│   └── utils.py             # Utility functions
├── tests/
│   └── test_ocr.py          # Test files
├── pyproject.toml           # Project configuration
├── LICENSE                  # MIT License
└── README.md                # Project documentation

Architecture

This project uses a subprogram parallel architecture, with 4 recognition tools that are independent and can be called separately:

┌─────────────────────────────────────────────────────────┐
│                   MCP Server (Dispatch Layer)            │
├──────────────┬──────────────┬─────────────┬─────────────┤
│ Subprogram 1 │ Subprogram 2 │ Subprogram 3│ Subprogram 4│
│ocr_recognize │  ocr_table   │ocr_handwrite│ ocr_formula │
│Text Recognition│Table Recognition│Handwriting Recognition│Formula Recognition│
└──────────────┴──────────────┴─────────────┴─────────────┘
       ↓              ↓              ↓              ↓
   OCREngine     TableParser    OCREngine    FormulaEngine

Features:

• Each subprogram runs independently without dependencies

• Supports parallel calls, can process multiple tasks simultaneously

• Creates independent instances for each call to avoid state conflicts

• Can individually extend or replace any subprogram

Dependencies

• PaddleOCR: Core OCR recognition engine

• PaddlePaddle: Deep learning framework

• OpenCV: Image processing

• MCP: Model Context Protocol service framework

License

This project is open-sourced under the MIT License.

Related Links

• PaddleOCR

• Model Context Protocol