OCR

The MCP server binds to native Tesseract libraries, which must be installed at the system level before building the Node.js addon. Install Tesseract and the English language data using your system package manager.

# macOS
brew install tesseract
brew install tesseract-lang  # all language packs

# Ubuntu/Debian
sudo apt install tesseract-ocr libtesseract-dev libleptonica-dev

# Windows (Chocolatey)
choco install tesseract

Clone the MCP OCR Server repository to your local machine.

git clone https://github.com/dangvinh/mcp-ocr-server.git
cd mcp-ocr-server

Create a tessdata directory and download the trained data for English (and any other languages you need). The TESSDATA_PREFIX environment variable tells Tesseract where to find these files.

mkdir -p tessdata
wget https://github.com/tesseract-ocr/tessdata/raw/main/eng.traineddata -P tessdata/

# Optional: add more languages
wget https://github.com/tesseract-ocr/tessdata/raw/main/fra.traineddata -P tessdata/
wget https://github.com/tesseract-ocr/tessdata/raw/main/deu.traineddata -P tessdata/

Compile the native C++ OCR library and then build the Node.js addon that wraps it. Both steps are required — the addon depends on the compiled core library.

cd cpp
npm run build-core    # Compiles libmcp_ocr.a via CMake
npm run build-addon   # Builds ocr_addon.node via node-gyp

Set TESSDATA_PREFIX to point to your tessdata directory and OCR_LANG to the default recognition language. These can be set in your shell profile or passed directly in the MCP client config.

export TESSDATA_PREFIX=/path/to/mcp-ocr-server/tessdata
export OCR_LANG=eng

Add the OCR MCP server to your Claude Desktop configuration. Pass the TESSDATA_PREFIX and OCR_LANG environment variables so the native Tesseract library can locate its language data.

{
  "mcpServers": {
    "mcp-ocr": {
      "command": "node",
      "args": ["/path/to/mcp-ocr-server/index.js"],
      "env": {
        "TESSDATA_PREFIX": "/path/to/mcp-ocr-server/tessdata",
        "OCR_LANG": "eng"
      }
    }
  }
}

Verify the build succeeded by running the ocr_cli executable directly on a test image before connecting it to an AI client.

TESSDATA_PREFIX=./tessdata ./cpp/ocr_cli path/to/test-image.png