The integration relies on HAS to expose the MCP endpoint, GraphQL/Web API to access Hopex data, and standard Hopex mechanisms for rights, connection profile, and repository context.

The MCP Server does not add a separate data governance layer. It exposes capabilities already controlled by the repository, the GraphQL schema, and HAS configuration.

The MCP endpoint exposed by HAS is the main entry point. It is published over HTTPS, like the other services exposed by Hopex Application Server, and is reachable from the AI client when the environment network, security, and authentication rules allow it.

The standard endpoint format is https://<hopex-host>/mcp. Replace <hopex-host> with the actual Hopex instance domain exposed to the AI client.

The endpoint URL is declared in the AI client when the MCP Server is added. Additional parameters, such as the authentication mode, x-api-key or x-schema headers, and client-specific options, are configured in the client according to its documentation.

Hopex capabilities are exposed to the AI client as MCP tools. These tools map to the capabilities available in the Hopex MCP Server implementation: schema introspection, query guidance, GraphQL query execution, authorized mutations, relationship exploration, object counting, and diagram access.

Tool selection is based on the exposed metadata and the user request. The MCP Server exposes the capabilities and enforces the authorized Hopex scope.

Each MCP call uses a Hopex context that includes authentication, GraphQL schema, and rights.

The Hopex user context identifies the user, repository, and connection profile associated with the call. Responses returned to the AI client remain aligned with this context.

The MCP Server associates exchanges with an MCP session. This session tracks the connected AI client, the schema used, the authentication mode, and the technical context of the call.

MCP sessions are monitored through HAS supervision. Session events make it possible to correlate calls, the identified Hopex user, executed tools, and loaded resources.

The integration model is identical in SaaS and on-premises environments: the AI client calls the MCP endpoint exposed by HAS, and the MCP Server accesses Hopex through authorized services.

Deployment differences are limited to the surrounding environment: published URL, identity provider, certificates, proxies, gateways, firewall rules, and administration constraints.