Expose Data Agent schema, question interpretation, and generated query through MCP

I would like to request additional context and metadata capabilities for the Microsoft Fabric Data Agent MCP server.

It would be very useful if the MCP server could expose more information about the Data Agent itself and how it processes a user's question.

1. Expose the Data Agent data schema

It would be useful to retrieve through MCP the data schema available to the Data Agent, including information such as:

• Tables or semantic model entities
• Columns and fields
• Data types
• Relationships
• Measures
• Descriptions and metadata
• Other relevant schema information available to the Data Agent

Ideally, this information could be exposed through a dedicated MCP resource or tool.

This would allow an MCP client to understand what data the Data Agent has access to without having to maintain a separate copy of the schema.

2. Expose the interpreted or reformulated question

It would also be very useful to retrieve information about how the Data Agent understood the user's question.

For example:

Original question:
"How many customers did we lose last month?"

Reformulated / interpreted question:
"Calculate the number of customers who became inactive during the previous calendar month."

If available, it would also be useful to expose structured information such as:

• Original user question
• Reformulated or normalized question
• Detected intent
• Relevant tables or entities
• Relevant columns or measures
• Filters and conditions identified from the question

This would help developers understand how the Data Agent interpreted the request.

3. Expose the generated and executed query

It would also be very valuable to expose the query generated by the Data Agent to answer the user's question.

Depending on the underlying data source, this could include:

• SQL
• DAX
• KQL
• Other query languages supported by Fabric Data Agents

For example, the MCP response or a dedicated MCP resource could expose:

Query language:
SQL

Generated query:
SELECT COUNT(*) FROM Customers WHERE ...

Or:

Query language:
DAX

Generated query:
CALCULATE(...)

It would also be useful to know:

• Which query language was selected
• The generated query
• The final query that was actually executed, if it differs
• The data source or semantic model targeted
• The tables, measures, or entities used by the query
• Execution status
• Query execution duration, when available
• Query errors or warnings, when applicable

This information would be extremely useful for debugging, auditing, observability, and explaining how the Data Agent produced its answer.

Why this would be useful

Ideally, an MCP client could retrieve a structured execution context such as:

Original question
→ Reformulated / interpreted question
→ Relevant schema and entities
→ Selected query language (SQL / DAX / KQL / etc.)
→ Generated query
→ Executed query
→ Result
→ Final Data Agent answer

The goal is not to expose the model's private internal reasoning, but rather to expose the structured interpretation, generated query, schema context, and execution metadata that can safely be shared with the MCP client.

These capabilities would make the Fabric Data Agent MCP server much more useful for building transparent, debuggable, auditable, and context-aware applications.

Could Microsoft please consider exposing the Data Agent schema, structured question interpretation, and generated/executed queries through the MCP server?