The Privacy Protector: MCP-based LLM agent for Health Support

Large Language Model (LLM) is increasingly utilized for healthcare support. However, at present, the usages and services suffer two limitations: firstly, some users' sensitive information should be better protected; and secondly, real time dynamic information should be incorporated for updated support. This paper proposes two new methods to address these limitations. Firstly, a privacy protection method is integrated into the use of LLM. It designs and implements encryption and substitution techniques to protect sensitive data while preserving the necessary information for LLM to provide advice. Our experimental results show that our proposed method achieves a 58.3% reduction in sensitive information leakage. Additionally, we propose a LLM agent augmented with MCP servers, so that the most recent data and information can be added into LLM for updated advice and answers. It demonstrates significantly enhanced capabilities to access external tools, such as real-Time weather APIs. Therefore, it results in a 9.5-point improvement in response quality. Collectively, our proposed methods contribute to more sophisticated healthcare assistance.