9. Supervision and Observability
Hopex MCP observability relies on HAS supervision. The MCP Server emits technical events for sessions, identified users, tool or resource execution, errors, and returned volumes.
Supervision events contain operational metadata only. They do not include prompts, LLM responses, full GraphQL results, secrets, tokens, or API keys.
9.1. Supervision events
The MCP Server emits these main events:
|
Event
|
Purpose
|
Main metadata
|
|---|---|---|
|
MCPServerSessionStart / MCPServerSessionEnd
|
Trace session opening and closing.
|
Session ID, client name and version, schema, authentication mode, mutation flag, OAuth client or identity provider when applicable.
|
|
MCPServerUserIdentified
|
Link the MCP session to the Hopex user context.
|
Session ID, repository, login, person, and connection profile.
|
|
MCPServerToolExecution
|
Trace tool execution.
|
Session ID, tool name, execution status, result length, and estimated tokens.
|
|
MCPServerResourceExecution
|
Trace resource loading.
|
Session ID, resource name, execution status, result length, and estimated tokens.
|
9.2. Event correlation
The MCP session identifier correlates all events for the same session. It links client information, authentication mode, schema, Hopex user context, executed tools, loaded resources, and errors.
This correlation identifies whether an issue is related to client configuration, authentication, schema, rights, or mutation restrictions.
9.3. Usage monitoring
Session and execution events provide a technical view of MCP usage: connected AI clients, schemas, authentication modes, executed capabilities, errors, and estimated returned volumes.
9.4. Supervised data limits
Supervision is designed for operation and support. It traces the technical context without storing business content exchanged through the AI client.