The MCP Server exposes repository data within a controlled context: GraphQL schema, Hopex rights, connection profile, target repository, and MCP configuration.

The AI client does not have unrestricted repository access. It calls the tools exposed by the MCP Server and receives only the information authorized for the current scenario.

Accessible data corresponds to the objects and relationships exposed by the GraphQL schema used by the MCP Server. The scope covers the main Hopex domains: applications, business capabilities, processes, technologies, projects, and their relationships.

The EA Core schema provides a simplified view of the Hopex model for MCP usage and AI integrations. It helps the LLM use repository data while remaining aligned with the metamodel and available rights. Customer-specific metamodel extensions should follow the exposure principles defined in section 5.2 Customized metamodel exposure.

The exposed scope is defined by the GraphQL schema, connection profile, Hopex rights, and MCP Server configuration.

Access remains mediated by Hopex services. The connection profile defines the visible objects, properties, and relationships available through the MCP tools. For customized metamodels, the GraphQL scope should be reviewed against the principles in section 5.2 Customized metamodel exposure.

Each MCP call is attached to a Hopex context: repository, connection profile, GraphQL schema, and authentication mode. With OAuth 2.0, the repository and profile are carried by the validated OAuth context.

This context places the request in the correct Hopex environment. Responses returned to the AI client match the authorized repository, profile, and scope.

The MCP Server uses GraphQL capabilities to filter results and limit the returned volume. Queries can include filters, pagination, and result limits.

This keeps responses focused and limits unnecessary exposure when the repository contains many objects or relationships.

The MCP Server acts as a controlled access layer between Hopex and the AI client. It does not provide unrestricted repository access or direct repository access. The data returned to the AI client is limited by the selected GraphQL schema, Hopex rights, connection profile, MCP configuration, and call parameters.

After data is returned to the AI client, it may be processed by the LLM used by that client. This processing is outside the MCP Server scope and depends on the AI client deployment model, the selected LLM provider, the tenant configuration, and the applicable contractual terms.

Before connecting the MCP Server to an AI client using a public or third-party LLM, the customer must review how the provider handles received data and ensure this is consistent with its confidentiality, data protection, and AI usage policies.

Bizzdesign is not responsible for data exposure, storage, reuse, training, or processing performed by an AI client or by a public or third-party LLM provider selected, configured, or authorized by the customer.

MCP calls are logged for diagnosis, support, and security analysis. Logs identify the call context, endpoint used, validation result, and any errors.

Log retention follows the customer environment rules and Hopex supervision practices. Logs do not expose secrets, tokens, API keys, or sensitive data that is not needed for diagnosis.