MCP Mistral OCR

By everaldo GitHub

Model Context Protocol (MCP) Server for Mistral OCR API

Overview

what is MCP Mistral OCR?

MCP Mistral OCR is an OCR server that utilizes Mistral AI's OCR API to process images and PDFs, enabling users to extract text from various file formats.

how to use MCP Mistral OCR?

To use MCP Mistral OCR, you can either run it in a Docker container or set it up for local development. You need to provide your Mistral API key and specify the directory for local file processing.

key features of MCP Mistral OCR?

Processes local files (images and PDFs) and files from URLs.
Supports multiple file formats including JPG, PNG, and PDF.
Outputs results in JSON format with timestamps.
Docker containerization for easy deployment.
UV package management for local development.

use cases of MCP Mistral OCR?

Extracting text from scanned documents.
Processing images for text recognition.
Automating data entry from PDFs.

FAQ from MCP Mistral OCR?

What file types does MCP Mistral OCR support?

It supports images (JPG, PNG, etc.) and documents (PDF).

Is there a limit on file size?

Yes, the maximum file size is 50MB as enforced by the Mistral API.

How do I run MCP Mistral OCR in Docker?

You can build the Docker image and run it with your Mistral API key and local file directory.

Content

MCP Mistral OCR

An MCP server that provides OCR capabilities using Mistral AI's OCR API. This server can process both local files and URLs, supporting images and PDFs.

Features

Process local files (images and PDFs) using Mistral's OCR
Process files from URLs with explicit file type specification
Support for multiple file formats (JPG, PNG, PDF, etc.)
Results saved as JSON files with timestamps
Docker containerization
UV package management

Environment Variables

MISTRAL_API_KEY: Your Mistral AI API key
OCR_DIR: Directory path for local file processing. Inside the container, this is always mapped to /data/ocr

Installation

Installing via Smithery

To install Mistral OCR for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @everaldo/mcp/mistral-crosswalk --client claude

Using Docker

Build the Docker image:

docker build -t mcp-mistral-ocr .

Run the container:

docker run -e MISTRAL_API_KEY=your_api_key -e OCR_DIR=/data/ocr -v /path/to/local/files:/data/ocr mcp-mistral-ocr

Local Development

Install UV package manager:

pip install uv

Create and activate virtual environment:

uv venv
source .venv/bin/activate  # On Unix
# or
.venv\Scripts\activate  # On Windows

Install dependencies:

uv pip install .

Claude Desktop Configuration

Add this configuration to your claude_desktop_config.json:

{
  "mcpServers": {
    "mistral-ocr": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MISTRAL_API_KEY",
        "-e",
        "OCR_DIR",
        "-v",
        "C:/path/to/your/files:/data/ocr",
        "mcp-mistral-ocr:latest"
      ],
      "env": {
        "MISTRAL_API_KEY": "<YOUR_MISTRAL_API_KEY>",
        "OCR_DIR": "C:/path/to/your/files"
      }
    }
  }
}

Available Tools

1. process_local_file

Process a file from the configured OCR_DIR directory.

{
    "name": "process_local_file",
    "arguments": {
        "filename": "document.pdf"
    }
}

2. process_url_file

Process a file from a URL. Requires explicit file type specification.

{
    "name": "process_url_file",
    "arguments": {
        "url": "https://example.com/document",
        "file_type": "image"  // or "pdf"
    }
}

Output

OCR results are saved in JSON format in the output directory inside OCR_DIR. Each result file is named using the following format:

For local files: {original_filename}_{timestamp}.json
For URLs: {url_filename}_{timestamp}.json or url_document_{timestamp}.json if no filename is found in the URL

The timestamp format is YYYYMMDD_HHMMSS.

Supported File Types

Images: JPG, JPEG, PNG, GIF, WebP
Documents: PDF and other document formats supported by Mistral OCR

Limitations

Maximum file size: 50MB (enforced by Mistral API)
Maximum document pages: 1000 (enforced by Mistral API)

No tools information available.

No content found.