Summary
Local OCR with no key and no upload, and three input shapes so you rarely have to convert anything first.
Taking a URL directly is the detail that saves steps — no downloading an image just to hand over a path. Because it is Tesseract underneath, quality tracks Tesseract: clean scans read well, and the language pack you have installed decides what it can recognise, which is what get_supported_languages is for.
What it is
An OCR server over Tesseract that takes an image from three different sources and returns the text it finds.
What you get
- perform_ocr accepts a local file path, an image URL, or raw image bytes
- get_supported_languages lists the OCR languages available on your install
- Tesseract installed automatically on macOS via Homebrew and on Linux via apt, dnf or pacman; Windows gets manual instructions
- Errors handled rather than swallowed
Requirements
Pip install mcp-ocr, or uv pip install mcp-ocr. Tesseract is pulled in for you on supported platforms. The client launches it as python -m mcp_ocr. No account and no API key — recognition runs locally, so images never leave the machine.
Setup effort
One command — pip install mcp-ocr