MCP Engine Semantic Authoring

PBI Semantic Authoring

Semantic-layer work through manage_semantic. Read only the references needed for the current task.

Decide the semantic path

1. Prefer a measure for reusable business logic. Choose a calculated column only when a persisted row-level attribute is required — creating one is a manage_schema create_calc_column operation covered by mcp-engine-schema-authoring.
2. Ground names first: list_model { "operation": "list", "spec": { "type": "measures" } } (repeat with "tables"; spec.type takes one value per call), and check for an existing equivalent with { "operation": "search", "spec": { "query": "<name>", "mode": "name" } } before creating a near-duplicate measure.
3. Keep semantic edits separate from table or relationship edits unless the task explicitly spans both.

Branch by object type

• Measures and KPIs (create_measure, update_measure, create_kpi): measure-authoring-guide. When logic placement is unclear: modeling-best-practices-guide.
• Calculation groups (create_calc_group): calc-groups-guide. DAX UDFs (create_udf): dax-udf-guide.
• Named expressions (create_named_expression): named-expressions-guide. Power Query parameters (create_pq_parameter, RangeStart/RangeEnd): power-query-parameters-guide.
• Model properties (descriptions, culture, annotations, implicit-measure behavior): model-properties-guide. Naming, folders, and style cleanup across many objects: semantic-layer-style-guide.

Guardrails

• Follow the model's existing naming and folder conventions unless the user requests a deliberate change.
• Validate each new or changed measure with a scoped run_query { "operation": "execute", "query": "<query evaluating the measure>" } in a realistic filter context.
• Use dry_run: true and per-item identifiers for bulk semantic edits.

Report results

After semantic work, report:

1. Each object created or changed, with its DAX or M expression.
2. Whether dry_run or bulk mode was used and the outcome.
3. The validation query run and its result.
4. Naming or folder decisions that follow (or deliberately break) model conventions.