DAB's long-term direction for AI-powered data analytics: secure API layer or foundation for semantic querying and BI?

[Codology (Mourad57)]

Hello Team,

First, thanks for the great work on Data API Builder, especially the MCP integration!

I've been experimenting with DAB through Claude Code against SQL Server's AdventureWorks database, comparing it with solutions such as [WrenAI](https://github.com/Canner/WrenAI), with the goal of building a self-hosted AI-powered platform for natural-language querying across multiple databases, including complex BI scenarios.

So far, DAB has actually been faster and simpler to work with for many queries. However, its current MCP capabilities show limitations when handling more complex analytical questions.

For example, asking for "Revenue by product category and sales territory, including distinct customers, order counts, average order value, and revenue contribution percentage" requires several joins and aggregations that DAB's MCP tools cannot currently execute in a single operation.

I understand that views and stored procedures can encapsulate such complexity. However, relying on database-specific objects raises concerns about portability and maintainability across different SQL engines (SQL Server, PostgreSQL, MySQL, etc.), especially for a general-purpose AI querying platform.

I've also noticed promising roadmap items such as [#3208 (GraphQL queries through MCP)](#3208) and [#2840 (Persisted GraphQL documents)](#2840).

This brings me to a few architectural questions:

1. Complex querying and semantic capabilities

Is DAB expected to evolve toward more expressive, database-independent querying through MCP, supporting complex joins, aggregations, calculated measures, and potentially reusable business metrics?

Could it eventually incorporate some semantic-layer capabilities similar to WrenAI or Cube, or is this intentionally outside its scope?

2. Integration with data catalogs

Could DAB be complemented by platforms such as [DataHub](https://github.com/datahub-project/datahub) or [OpenMetadata](https://github.com/open-metadata/OpenMetadata), leveraging their metadata, relationships, business glossaries, and lineage to help AI agents understand and query databases more intelligently?

Is this an integration direction the team has considered?

3. Long-term architectural positioning

For a production-grade application where users ask natural-language questions of varying complexity across heterogeneous databases, would you consider DAB a suitable foundation, or recommend combining it with a dedicated semantic/query engine?

More fundamentally, is DAB's long-term vision to remain a secure, deterministic data-access layer (REST, GraphQL, MCP, RBAC, policies), or to progressively become a broader foundation for AI-driven analytical querying?

I'm not suggesting DAB should become a full BI engine. I'm mainly trying to understand its intended architectural boundaries and roadmap to make an informed long-term technology choice.

I'd really appreciate the team's perspective and any relevant roadmap discussions or recommended integration patterns.

Thanks!