6. Security
The security approach is designed to provide access to Hopex architecture knowledge through AI agents in a controlled manner aligned with enterprise security requirements. It protects architecture data, limits unnecessary exposure, and ensures that the MCP Server operates as a secure access layer rather than as an embedded AI solution.
Authentication and controlled access: Access to the MCP Server is protected through API key authentication, thereby helping ensure that repository knowledge can be queried only by authorized clients.Controlled exposure of repository content: The server provides access to repository content through a controlled interface, enabling architects to use architecture knowledge without unnecessarily exposing the underlying repository.Separation from AI model hosting: The MCP Server does not embed any AI agent or AI model. This separation preserves a clear boundary between the repository access layer and the AI runtime and provides organizations with greater flexibility in the governance and security of their AI environment.Organizations should also assess how architecture data is managed when AI agents rely on cloud-hosted LLMs. Even when the MCP Server provides controlled access to architecture knowledge, prompts and retrieved data sent to an external model may be processed outside the organization's direct technical perimeter. This creates governance concerns related to data exposure, retention, and the potential leakage of sensitive architecture information. Architects and security teams should therefore ensure that the selected AI configuration remains consistent with enterprise requirements for data handling, confidentiality, and approved cloud usage.
Security considerations may also vary depending on the tenant configuration and the selected LLM deployment model. Where the AI service is operated within a dedicated enterprise tenant, data management controls, access boundaries, logging, and compliance settings may differ materially from those associated with public or externally hosted LLM services. Consequently, the level of data exposure risk is determined not only by the MCP Server itself, but also by the tenant architecture, the cloud configuration, and the contractual and technical safeguards applied to the LLM environment.
Warning on data leakage risks: Sensitive architecture data should not be transmitted to external LLM services without appropriate safeguards, as prompts and retrieved content may result in the disclosure of confidential information beyond the organization's intended security boundary.