11. Tests and Validation
Validation covers server registration, connection, authentication, headers, simple read scenarios, errors, and supervision when a customer connects an AI client to the Hopex MCP Server.
Tests are performed from the AI client using simple prompts. The AI client triggers MCP calls in the background, based on the user prompt and, when applicable, LLM tool selection. The end user does not directly select MCP tools.
11.1. Integration preparation
Before testing, check that the AI client has the required connection data:
• Hopex MCP Server HTTPS URL
• Authentication mode: OAuth 2.0 with DCR or API key
• Optional headers: x-api-key, x-schema, x-allow-mutation
• Expected GraphQL schema or default GraphQL schema
• Network access from the AI client to the MCP Server
Secrets remain stored in the secure mechanism of the AI client.
11.2. MCP Server registration
Start by registering the MCP Server in the AI client. The client retrieves the metadata exposed by the Hopex MCP Server, including available tools and resources.
Expected result: the server is added without error, and the AI client records the declared capabilities. This check does not assess LLM behavior.
11.3. Connection and authentication
The AI client connects to the MCP endpoint exposed by HAS over HTTPS. This check validates network access, TLS certificate, any proxy, and the authentication mode.
With OAuth 2.0 and DCR, the AI client follows the authentication flow and obtains a token. In API key mode, it sends the x-api-key header when this mode is authorized on the server side.
11.4. Configuration header checks
When headers are configured in the AI client, the test checks their effect on the MCP Server. The AI client may not display transmitted headers, but the response confirms whether they are used.
x-schema: a read prompt produces a response consistent with the targeted GraphQL schema. Without this header, the default GraphQL schema is used.x-api-key: in machine-to-machine mode, the call is accepted when the transmitted key is valid and authorized on the server side.x-allow-mutation: in API key mode, this header is checked only for mutation scenarios.11.5. Simple read prompts
After the server is added, simple read prompts confirm that the AI client reaches the Hopex MCP Server and returns a readable answer.
Ask for the list or summary of types available in the repository.Search for a known Hopex object by name or identifier.Count objects of a given type.Ask for the main relationships of a known object.Ask for diagrams associated with a known object when diagrams exist.These prompts validate the integration path. They do not assess LLM reasoning quality. When a customized schema is exposed, validation should also include prompts covering the customized metamodel concepts described in section 5.2 Customized metamodel exposure.
11.6. Mutation tests
Mutation tests are performed only in a suitable environment with test data. They validate that write operations are accepted only when the server-side configuration and the client-side context allow them.
For API key mode, the test also validates the x-allow-mutation header. For irreversible deletion, the MCP Server requests explicit confirmation before execution.
11.7. Error handling tests
Errors are checked when they occur naturally during tests: connection, authentication, missing or invalid header, rejected mutation, or GraphQL error.
The MCP Server returns a structured response. The AI client then formats the message shown to the user.
11.8. Supervision checks
Tests from the AI client produce HAS supervision events. The check covers the MCP session, identified user, tools, resources, and errors.
These events link AI client activity to the MCP session, declared client, Hopex repository, connection profile, and executed processing.