FHE Image Processing Server

This project demonstrates a server-based application for applying image filters using Fully Homomorphic Encryption (FHE). It allows for secure image processing where the server never sees the unencrypted image content.

Overview

The system consists of two main components:

• FHE Server: Processes encrypted images without decrypting them
• MCP Server: Client interface that handles image preparation, encryption, and decryption

Features

• Apply various image filters to encrypted images
• Complete FHE workflow: encryption, processing, and decryption
• Process images while preserving privacy through homomorphic encryption
• RESTful API for client integration

Architecture

The system follows a client-server architecture:

1. The client sends an image URL to the MCP server
2. MCP server:
   • Downloads the image
   • Generates encryption keys
   • Encrypts the image
   • Sends the encrypted image to the FHE server
3. FHE server:
   • Processes the encrypted image without decryption
   • Returns the encrypted result
4. MCP server:
   • Stores the encrypted output
   • Can decrypt the result when requested

Setup

Prerequisites

• Python 3.11.4
• uv package installer (faster alternative to pip)

Installation

1. Clone the repository

2. Install MCP CLI:

   uv add "mcp[cli]"
   

3. Initialize filters:

   uv run generate_dev_files.py
   

4. Install MCP server to Claude Desktop:

   uv run mcp install mcp_server.py
   

   If uv run mcp install mcp_server.py doesn't work in your Claude Desktop, you can try to config bash {project_path}/mcp_server.exmaple.sh and replace {project_path} with your project path.

Usage

Starting the Servers

1. Start the FHE server:

   uv run fhe_server.py
   

   This will run on http://localhost:8000 by default

2. You can now access the MCP server through Claude Desktop

API Endpoints

FHE Server Endpoints:

• GET /: Welcome message
• GET /available_filters: List available filters
• POST /send_input: Upload encrypted image and evaluation key
• POST /run_fhe: Execute FHE computation
• POST /get_output: Retrieve encrypted output
• POST /fhe_full: Complete FHE workflow in a single request
• GET /test/image/{image_name}: Test endpoint for viewing images

MCP Server Tools:

• get_available_filters(): Get list of available FHE filters
• process_image_with_fhe(image_url, filter_name): Process an image through the FHE pipeline
• decrypt_output_image(user_id, filter_name, output_id): Decrypt processed image

Notes

• The system resizes images to 100x100 pixels for processing
• Encryption/decryption keys are generated per session
• Processed images are temporarily stored in the configured temp directories

Security Considerations

This demo implements FHE for educational purposes. In a production environment, additional security measures should be implemented:

• Secure key management
• HTTPS for all API communication
• Authentication and authorization
• Proper data retention policies