We have an AI system that uses Hibernate and Azure Databricks. How would you design the MCP server?

The MCP server should act as a mediator between the AI workload and the data platform. 1) API Gateway – expose REST/GraphQL endpoints that accept data transformation requests. 2) Service Layer – a stateless Java service that uses Hibernate to map domain objects to the relational store (Azure SQL or PostgreSQL). 3) Data Processing Layer – submit Spark jobs to Azure Databricks via the Databricks REST API; the service serializes input parameters, triggers the job, and streams results back. 4) Orchestration – use Azure Logic Apps or Apache Airflow to schedule batch jobs, manage dependencies, and handle retries. 5) Security – use Azure AD for authentication, store secrets in Azure Key Vault, and enforce network isolation via VNets. 6) Observability – log all requests to Azure Monitor, capture metrics (latency, throughput), and trace calls across services with OpenTelemetry. 7) Scalability – containerize the MCP service with Docker/Kubernetes (AKS) and autoscale based on CPU/memory or request queue depth. This design keeps the MCP server lightweight, leverages Hibernate for ORM, and offloads heavy analytics to Databricks, ensuring maintainability and performance.