Connect to a Power BI target and detect available engines.

Args: target_type: One of "pbi_desktop", "fabric_workspace", "pbip_folder", "pbix_file". target_ref: Reference to the target (path, workspace_id, etc.). auth_mode: "interactive" or "service_principal". tenant_id: Azure tenant ID (only used with service_principal).

Returns: ConnectResult with session_id, engines_available, and warnings.

Create a versionable plan from a template name + structured args.

Args: intent: One of the 4 MVP templates: safe_rename, audit, deploy, dax_regression. (Each template's required args go in options.) options: Template-specific args (e.g. for safe_rename: {"old_path": "T[A]", "new_path": "T[B]", "scope": "report_bindings"}). May also contain PlanOptions fields which are extracted before passing template args to the builder.

Returns: PlanResult with plan_id, plan_yaml, steps, risk_score, estimated_changes.

Execute a plan with automatic rollback on failure.

Args: plan_id: ID of the plan previously created by plan_change. dry_run: If True, execute steps without side effects (dry_run executor is used regardless of registered engines). confirm_each_step: Reserved for per-step elicitation in v2; accepted but not enforced in MVP.

Returns: ApplyResult with execution status, executed steps, rollback info.

Composite audit (BPA + DAX lint + WCAG) on a PBIP folder.

Pre-deploy gate + publish PBIP + configure refresh + initial refresh.

Trigger and optionally wait for a dataset refresh.

Run DAX queries vs a baseline JSON and diff results.

Diff two semantic models (PBIP folders or snapshots).

Evaluate findings against a pre-deploy gate profile.

Generate a Markdown data dictionary (with Mermaid ER diagram) for a PBIP.

Apply a colorblind-safe theme + backfill alt text + re-audit WCAG.

Add a DAX measure with mandatory lint validation.

Lint gate blocks writes if any finding has severity ≥ fail_on_severity (default: warning). dry_run=True returns findings without writing.

Scaffold a PBIR folder from an existing dataset.

Creates .Report/, theme.json, report.json, and a sample page. Never overwrites existing files (returns warnings instead).

Deterministic edit on a single visualContainer in a PBIR page.

Field-level merge: only the fields in the input change; unspecified fields are preserved. Atomic write via temp-then-rename.

Detect measures that share structure (e.g. X YTD/QTD/MTD) and consolidate into a calculation group.

Returns the plan + (optionally) writes the calc group if auto_apply=True.

For each KPI in the JSON list, return a primary visual + alternatives.

Uses the viz/visual_suggester for recommendation logic.

Synthesize a PBIR page from an NL brief.

Composes viz/visual_suggester + python_report for actual file I/O.

Heuristic performance analyzer for PBIR pages.

Reads pages/*/page.json and detects anti-patterns that impact perceived load (visual density, pie/donut, custom visuals, conditional-formatting). Returns a 0-100 score, estimated load in ms, and a per-page list of hotspots with low/medium/high cost classification and fix suggestions.

Heuristic qualitative auditor (hierarchy/density/narrative/mobile/cohesion).

Reads PBIR pages and scores 0-100 with category breakdown + per-finding suggestions. Strictness adjustable (lenient|standard|strict).

Best-effort screenshot capture of PBIR pages.

Without Power BI Desktop Bridge this emits SVG wireframes + JSON manifests (deterministic for regression diff). Real PNG/PDF needs superbi-mcp on Windows; absence is reported via rendering_warnings.

Generate a TMDL semantic model from a declarative spec.

Validates the spec (Pydantic + dangling-reference check + basic DAX lint) and either returns the validation result (dry_run=True) or atomically writes the PBIP layout (dry_run=False). Output contains tables_created, relationships_created, hierarchies_created and any lint_findings.

Apply RLS roles + members + run a test matrix against a TMDL model.

Validates the role spec, writes role <Name> blocks into definition.tmdl (atomic), and runs each test_query through the (injected) test_engine callable. Failures can roll back edits when rollback_on_test_failure is True.

Promote items between Fabric Deployment Pipeline stages.

Built-in gates (pre_deploy_check / audit_model_and_report / run_dax_regression) wrap the corresponding tools; plug custom gates via injected custom_gates. fabric_client is the injected REST adapter; absent it the tool runs in dry-run.

Snapshot a Fabric workspace into a local Git repo.

Reads workspace state via fabric_client.snapshot_workspace and commits each item as <type>/<name>.pbip. Warns on local uncommitted changes or large files (>50MB by default). dry_run skips the commit.

Deploy a local Git PBIP tree to a Fabric workspace.

For v3 inicial the only conflict-resolution mode is manual: conflicts are surfaced in items_skipped for human review. The optional pre_deploy_profile + pre_deploy_findings arguments can be passed via the input schema when you add them to the MCP wrapper.

Apply a Microsoft Purview sensitivity label to one or more items.

Batched via POST /admin/items/labels/bulkSet. Gated by an explicit *.Admin.* (or InformationProtectionPolicy.Apply.All) scope; missing scope elicits remediation. Names are SHA-256 redacted in outputs (toggle via redact_names).