MCP servers, also known as Model Context Protocol servers, are gaining traction as they facilitate interactions between Large Language Models (LLMs) and various external tools or services by translating natural language requests into structured operations. These servers extend LLM capabilities beyond their inherent knowledge by enabling real-time actions and maintaining a consistent interaction pattern with different tools. However, a critical aspect of their functionality—secrets management—is often overlooked, creating potential security risks. MCP servers need to juggle sensitive credentials like API keys and database credentials, which necessitates best practices for secure management, such as avoiding hardcoded secrets, implementing ephemeral credentials, securely injecting secrets in pre-built servers, and isolating access between different server instances. Infisical provides a solution to manage these secrets effectively, offering both self-hosted and managed cloud services, ensuring a robust security framework that supports the advanced functionalities of LLMs.