2. Hopex MCP Architecture
The Hopex MCP Server is delivered with the Hopex bundle and runs as a HAS module. It uses the standard HAS mechanisms for communication, security, deployment, and supervision.
No separate MCP Server package is required. The module is started from the HAS console, and the deployed MCP Server version is the one delivered with the installed Hopex bundle.
It does not access the Hopex repository directly. All requests go through the MCP endpoint exposed by HAS and are constrained by Hopex rights, the selected GraphQL schema, and MCP Server settings.
Hopex remains the system of record. The MCP Server exposes only authorized capabilities; the AI client uses the returned data to produce the user-facing answer.
Security follows the same principles as other applications consuming the Hopex Web API. MCP access is governed by Hopex rights, the selected GraphQL schema, the configured authentication mode, and the customer environment security rules.
2.1. Role in the Hopex ecosystem
The MCP Server lets an AI client query Hopex through an MCP-compatible interface, without exposing the repository or bypassing repository governance.
The information exposed through MCP depends on the GraphQL schema associated with the server. This schema defines the objects and relationships that can be accessed by the AI client. When the schema includes customer-specific metamodel extensions, the additional guidance in section 5.2 Customized metamodel exposure applies.
MCP is not an AI engine and does not replace Hopex. It provides an integration interface between the AI client and the repository.
MCP does not connect Hopex to any LLM by itself. The connection to an LLM depends entirely on the AI client selected, configured, and authorized by the customer.
Authentication and connection modes depend on the MCP Server configuration, the Hopex environment, and the AI client capabilities. The applicable modes and parameters are described in the security and configuration sections of this document.
2.2. Components
The main components are:
AI client: user-facing application used to submit a request and display the returned answer.MCP Server: HAS module exposing authorized MCP tools and providing the interface between the AI client and Hopex.HAS: runtime framework for the MCP Server. It provides communication, security, supervision, deployment, endpoint exposure, and start/stop of the MCP module from the HAS console. The deployed MCP Server version is aligned with the installed Hopex bundle.GraphQL/Web API: access layer for Hopex objects and relationships exposed by the configured schema.Hopex repository: system of record for architecture data, access rights, and business consistency.2.3. Communication flow
Communication uses HTTPS across the exposed flow. The AI client calls the MCP endpoint exposed by HAS, while the MCP Server accesses Hopex through GraphQL/Web API.
AI client to MCP Server: the MCP-compatible AI client calls the MCP endpoint exposed by HAS over HTTPS. The exchanged messages follow the MCP protocol supported by the installed Hopex MCP Server version.Secure transport: HTTPS is used across the exposed communication path. Depending on the customer architecture, traffic may pass through TLS termination, a reverse proxy, a gateway, firewall rules, and certificates.MCP Server in HAS: the server receives the request, applies the HAS execution and security framework, and prepares access to the authorized Hopex capabilities.MCP Server to Hopex: the MCP Server accesses Hopex data through GraphQL/Web API, also within the HTTPS framework exposed by HAS. The configured GraphQL schema determines the accessible objects and relationships.Response to the AI client: Hopex returns authorized data to the MCP Server, which sends a structured response back to the AI client. The AI client then presents the information to the user.2.4. Responsibilities
Responsibilities are separated: the AI client handles interaction, the MCP Server exposes authorized Hopex tools, HAS provides runtime and supervision, and Hopex manages repository data and rights.
Hopex provides the MCP endpoint, authentication, controlled data access, and the GraphQL schema based on the configured metamodel. AI agents, LLM selection, prompting strategies, and customer-specific schema adaptation remain outside its scope.
2.5. Architecture diagram
