aggregate_records should default fields to *

[souvikghosh04]

Why make this change?

• Closes [Enh]: aggregate_records should default fields to * #3280
• The aggregate_records MCP tool requires callers to explicitly pass a field argument, even for the most common case (function: count) where * is the only sensible value. The issue requests defaulting field to "*" in the JSON Schema so LLM clients (and other MCP callers) can omit the argument for count-all scenarios, aligning the tool with the conventional COUNT(*) shape.

What is this change?

• Updated the aggregate_records tool's input schema in AggregateRecordsTool.cs:
  • Added "default": "*" to the field property and clarified its description.
  • Removed "field" from the schema's required array (now only entity and function are required).
• Updated argument parsing in TryParseAndValidateArguments so that when field is omitted:
  • function: count ⇒ field defaults to "*" (existing count(*) validation/path is preserved).
  • function: avg | sum | min | max ⇒ returns an InvalidArguments error explaining a specific numeric field is required, since "*" is invalid for those aggregations.
• Existing behavior is preserved when field is explicitly supplied (including the field='*' + non-count rejection and the distinct + count(*) rejection).

How was this tested?

• Integration Tests
• Unit Tests
  • Updated GetToolMetadata_HasRequiredSchemaProperties to assert:
    • field is no longer in the required array.
    • The field property declares "default": "*".
  • Updated AggregateRecords_MissingOrInvalidRequiredArgs_ReturnsInvalidArguments data rows to:
    • Remove the now-invalid "missing field with count" failure case.
    • Add new rows asserting avg / sum without field return an InvalidArguments error mentioning field.
  • All 60 tests in AggregateRecordsToolTests pass locally.

Sample Request(s)

Before this change (required):

{
  "tool": "aggregate_records",
  "arguments": {
    "entity": "Book",
    "function": "count",
    "field": "*"
  }
}

After this change (field may be omitted for count):

{
  "tool": "aggregate_records",
  "arguments": {
    "entity": "Book",
    "function": "count"
  }
}

Non-count functions still require an explicit field:

// Returns InvalidArguments: "Missing required argument 'field'. The default value '*'
// is only valid with function 'count'; for function 'avg', provide a specific numeric field name."
{
  "tool": "aggregate_records",
  "arguments": {
    "entity": "Book",
    "function": "avg"
  }
}

[Copilot]

Pull request overview

This PR updates the aggregate_records built-in MCP tool to make field optional for the common count(*) scenario by declaring a JSON Schema default ("*"), removing field from the required list, and adding runtime validation to only apply that default for function=count.

Changes:

• Updated aggregate_records input schema: field is no longer required and declares "default": "*".
• Updated argument parsing to default field to "*" only when function=count; non-count functions now error if field is omitted.
• Updated unit tests to validate the schema’s required list and field default, and to adjust missing-field validation cases.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/Azure.DataApiBuilder.Mcp/BuiltInTools/AggregateRecordsTool.cs	Makes field optional in the schema and implements defaulting/validation behavior in argument parsing.
src/Service.Tests/Mcp/AggregateRecordsToolTests.cs	Updates schema assertions and missing-argument test cases to reflect the new optional field behavior.