Nanonets MCP Server

How to use

You interact with the MCP server through an MCP client. The server provides endpoints to convert individual images, entire PDFs, Word documents, and Excel workbooks into structured markdown. Use the appropriate tool for your input type and rely on the server to preserve layout elements like tables, headings, and multi-page boundaries.

Common usage patterns include converting a single image to markdown, processing a multi-page PDF document with page separators, extracting text from Word or Excel files, and querying the server for a full document representation. When you pass a document or image, the server returns a structured markdown representation that captures content structure and formatting.

Key capabilities you can rely on include: recognizing text and paragraphs, preserving table structures, extracting LaTeX equations, describing images, detecting signatures and watermarks, identifying checkboxes, handling complex layouts, and properly separating multi-page documents.

How to install

Prerequisites you need before installation: Python ≥ 3.10, a working container runtime if you plan to use Docker, and MCP ≥ 1.0.0. Optional dependencies exist for enhanced PDF, Word, and Excel support.

Option 1: Docker (Recommended with GPU)

# Clone the repository
git clone <repository-url>
cd nanonets_mcp

# Build and run with Docker Compose (requires NVIDIA Docker runtime)
docker-compose up --build

Prerequisites for GPU support: NVIDIA GPU with CUDA support, NVIDIA Docker runtime installed, Docker Compose v3.8+.

Option 2: Local Installation

# Clone the repository
git clone <repository-url>
cd nanonets_mcp

# Install dependencies with uv
uv pip install -e .

Starting the server

Choose the method you used during installation to start the MCP server. If you used Docker, start the container; if you installed locally, run the Python module that launches the server.

Starting with Docker

# Start with Docker Compose
docker-compose up

# Or run directly with Docker
docker run --gpus all -p 8000:8000 nanonets-mcp:latest

Starting locally

# Start the MCP server
nanonets-mcp

# Or run directly
python -m nanonets_mcp.server

Integration example with Claude Desktop

Connect your Claude Desktop configuration to the MCP server to enable OCR workflows directly from Claude. Use the provided MCP server entry.

{
  "mcpServers": {
    "nanonets-ocr": {
      "command": "nanonets-mcp"
    }
  }
}

ocr_image_to_markdown

Converts an input image to a structured markdown representation, preserving layout details such as text blocks, headings, and simple tables.

ocr_pdf_to_markdown

Converts an entire PDF document to structured markdown, including page separators and preserved page content order.

process_word_to_markdown

Converts a Word (.docx) document to structured markdown, preserving headings, paragraphs, and tables.

process_excel_to_markdown

Converts an Excel workbook to structured markdown, preserving worksheets and table data.

get_supported_formats

Provides information about supported input formats and processing capabilities.