Mcp Ocr

What is Mcp Ocr?

Mcp Ocr is a production-grade OCR server built using the Model Context Protocol (MCP) that extracts text from images via Tesseract OCR. It is intended for developers who need OCR capabilities integrated into MCP-compatible tools like Claude for Desktop.

How to use Mcp Ocr?

Install with pip install mcp-ocr or uv pip install mcp-ocr, then start the server with python -m mcp_ocr. To use with Claude for Desktop, add an entry named "ocr" in the mcpServers section of the configuration file, specifying the same command. The server exposes two tools: perform_ocr (for extracting text from files, URLs, or raw bytes) and get_supported_languages (to list available OCR languages).

Key features of Mcp Ocr

• Extracts text from images using Tesseract OCR
• Supports local files, image URLs, and raw bytes
• Automatic Tesseract installation on macOS, Linux, and Windows
• Multi-language OCR support
• Production-ready error handling

Use cases of Mcp Ocr

• Extract text from scanned documents or screenshots
• Read text from images hosted on the web
• Integrate OCR into MCP-based automation workflows
• Retrieve supported languages for multilingual image processing

FAQ from Mcp Ocr

What OCR engine does Mcp Ocr use?

It uses Tesseract OCR, an open-source optical character recognition engine.

On which platforms is Tesseract installed automatically?

Automatic installation is supported on macOS (via Homebrew), Linux (via apt, dnf, or pacman), and Windows (with manual installation instructions provided).

Can I specify the OCR language?

Yes, Mcp Ocr supports multiple languages. Use the get_supported_languages tool to list available languages, and pass the desired language code to perform_ocr (though the README does not explicitly show the parameter, the feature is stated).

How do I pass raw image bytes to the OCR tool?

Call perform_ocr with the bytes object as the argument; the server handles extraction from raw bytes.

What runtime dependencies are required?

Python and Tesseract OCR. The server is installed via pip or uv and runs as an MCP server over stdio (no authentication mentioned).