deeppowers

# DeepPowers <div align="center"> <a href="deeppowers.xyz"> <img src="https://github.com/deeppowers/deeppowers/blob/main/assets/deeppowers_logo.jpg" style="margin: 15px; max-width: 300px" width="30%" alt="Logo"> </a> </div> <p align="center"> <em>DEEPPOWERS is a Fully Homomorphic Encryption (FHE) collaboration framework built for MCP (Model Context Protocol), aiming to provide end-to-end privacy protection and highly efficient computation for the upstream and downstream ecosystems of the MCP protocol. By deeply integrating the FHE framework with the MCP protocol, we are dedicated to creating a secure, efficient, and scalable computing framework for MCP. This ensures that data remains encrypted throughout transmission, storage, and computation, while also supporting complex computing logic, eliminating unnecessary data transmission and computation, and thereby rapidly improving MCP's operational efficiency.</em> </p> <p> <em align="center"> By removing delays in MCP interactions, it aims to provide robust momentum for the Model Context Protocol (MCP) ecosystem. DEEPPOWERS unleashes a higher level of efficiency, collaboration, and performance for MCP workflows. Support for various MCP servers and leading large language models (LLMs) such as DeepSeek, GPT, Gemini, and Claude ensures unparalleled versatility and enhanced collaborative efficiency. </em> </p> [](LICENSE) [](docs/userguide.md) [](https://www.python.org/) [](https://en.cppreference.com/w/cpp/17) ## Overview Fully Homomorphic Encryption (FHE) allows computations (such as addition, multiplication, etc.) to be performed directly on encrypted data without decryption. The computation results remain encrypted, and only authorized users can decrypt them. FHE resolves the conflict between data privacy and computational efficiency and is suitable for scenarios such as cloud computing, medical data analysis, and financial transactions. Its core lies in ensuring data remains encrypted throughout the process, eliminating the risk of privacy leakage in intermediate steps, while supporting complex computations, providing the ultimate guarantee for data security and compliance. It supports languages such as C++, Python, and CUDA, facilitating integration into the existing MCP ecosystem. ## Key features - **End-to-end Encryption**: Complete protection for user-LLM interactions with data remaining encrypted throughout the entire process - **Secure Task Execution**: Encrypted data transmission and computation for sensitive operations without exposure to third parties - **Seamless MCP Integration**: Deep integration with the Model Context Protocol ecosystem, enhancing MCP's capabilities with privacy-preserving computations - **Accelerated MCP Workflows**: Process encrypted data locally and reduce unnecessary transfers, significantly improving MCP operational efficiency - **Enhanced Tool Security**: Enable secure execution of MCP tools that handle sensitive data, expanding the scope of privacy-preserving tasks - **Cross-Model Compatibility**: Work with various MCP servers and LLM providers including DeepSeek, GPT, Gemini, and Claude - **Reduced Latency**: Minimize delays in MCP interactions by optimizing the encryption, computation, and decryption processes - **Privacy-Preserving AI**: Complete AI assistance workflows while maintaining strict data privacy throughout - **Scalable Architecture**: Designed to grow with the MCP ecosystem, supporting emerging models and computational tools ## Powered by Concrete-ML DeepPowers is built on [Concrete-ML](https://github.com/zama-ai/concrete-ml), an open-source library developed by Zama for privacy-preserving machine learning using Fully Homomorphic Encryption. Concrete-ML provides: - Privacy-preserving ML framework using FHE - Compatibility with traditional ML frameworks (scikit-learn, PyTorch) - Tools for quantizing and converting ML models for FHE execution - Simple and intuitive APIs similar to scikit-learn With Concrete-ML as its foundation, DeepPowers extends these capabilities to LLM interactions, creating a secure environment for AI-assisted tasks. ## Structure  ### 1. MCP Client - **Responsibilities:** - Local FHE encryption/decryption - Key pair generation and management (private key never leaves the client) - Task submission and encrypted result display - User authentication and permission control ### 2. MCP Server - **Responsibilities:** - Secure storage and dispatching of encrypted data and tasks - Task management and scheduling - Node status monitoring and logging - No access to plaintext data ### 3. TEE Compute Node - **Responsibilities:** - Executes encrypted computa