Delete a relationship by name or by endpoint columns.

Update properties of an existing relationship (cardinality, direction, is_active, name).

Delete a table. Removes associated relationships and measures.

Delete a column from a table.

Rename a table. Callers are responsible for updating dependent DAX expressions.

Rename a column. Callers are responsible for updating dependent DAX.

Rename a DAX measure. Callers must update downstream DAX expressions themselves.

Reload cached TOM schema from the server (cheaper than full reconnect).

Parse-check a DAX expression by running a zero/one-row probe and catching errors.

kind='scalar' wraps the expression with EVALUATE ROW("v", ). kind='table' wraps the expression with EVALUATE TOPN(0, ).

semantic=True additionally checks every Table[Column] / [Measure] reference against the live model index and heuristically flags a percent-shaped format_string on a scalar-money expression — the response then carries a semantic block (see pbi_validate_dax_semantic_tool).

Detect missing, ambiguous, bidirectional, and orphaned model structures.

Validate measures, formats, and measure/column name collisions.

Detect table, column, and measure name collisions before writes.

Detect text columns that look like dirty dates.

Execute lightweight DAX probes to detect visuals with no data.

Detect cycles in the measure dependency graph.

Builds a graph of measure → referenced measures (parsed from DAX [Name] tokens that match a known measure name) and runs a DFS to find strongly connected cycles. Self-references are reported separately.

Detect required visuals that are absent from a page.

Each requirement entry is a dict with at least visual_type and optionally count (default 1), contains_field (a reference like Date.Year that must appear in the visual's prototypeQuery), and label (free-form name surfaced in the report).

Generate a Markdown correction report aggregating all analysis tools.

Output sections: model overview, star-schema verdict, circular dependency scan, optional rubric scoring, audit issues. Writes to output_path and returns the path plus an inline summary.

Export model, DAX, layout, binding, and score validation as JSON.

Aggregate scoring across multiple validators.

Each criterion is a dict with:

• id (str): unique identifier

• label (str): human description

• check (str): one of star_schema, no_circular_deps, power_query_steps, missing_visuals, measure_exists

• weight (float, default 1.0)

• params (dict): check-specific parameters

Returns per-criterion verdicts plus a weighted total score in [0, 1].

Verify that a Power Query (M) expression contains expected step patterns.

Each entry in expected_steps is treated as a substring (or regex if it starts with re:) that must appear at least once in the M expression. Useful for grading exercises: e.g. checking that a postal-code column has been left-padded to 5 chars, or that rows with null customer ids are filtered out.

Confirm a model follows star schema topology.

Heuristic: a fact table participates as the many-side in ≥1 relationship, a dimension table participates as the one-side. Tables that are both are flagged as bridge tables. Direct dimension-to-dimension relationships are flagged as snowflake violations. Fact tables wired to other facts are flagged as fact-to-fact (constellation).

Optional fact_table_hints lets callers force-tag tables as facts (matched case-insensitively); useful when a fact table has no incoming relationships yet because the model is being built incrementally.

Validate a DAX boolean filter expression before visual probes.

Generate and execute smoke tests for DAX measures.

Validate relationship cardinality, direction, duplicates, and ambiguity before creation.

Detect overlaps, excessive whitespace, tiny visuals, and missing titles.

Optional knobs to silence noise on intentionally dense pages:

• ignore_warnings: list of warning types to drop (e.g. ["too_many_visuals", "visual_too_small", "missing_title", "excessive_whitespace", "layout_overloaded"]). Issues are never ignored — only warnings.

• only_pages: restrict the scan to a list of page display names / internal names. Use to lint just the new pages an LLM produced.

• max_visuals_per_page: override the default too_many_visuals threshold (defaults to MAX_VISUALS_PER_PAGE). Set high to effectively disable the warning without ignoring it.

Alias-level visual binding validation with clearer tool naming.

Score dashboard quality across model, DAX, layout, and readability.

Run a complete QA scenario against the active model and optional extracted layout.

Compare two extracted report versions by pages, visuals, and layout score.

Validate that a patched PBIX still contains a readable, persistent report layout.

Open a PBIX in Power BI Desktop and scan for visible repair-error signals.

use_windows_ocr defaults to False because the underlying screenshot captures the entire primary desktop, not just the Power BI window. The OCR helper now returns only the matched signal labels and a length — never raw recognized text — so screen contents from other applications cannot leak through the response.

Return calc-dependency graph rows (source → referenced object) from DISCOVER_CALC_DEPENDENCY.

List all model roles with their members and table filters.

Create or update a model role.

Delete a model role.

Set (or remove when filter_expression is None/empty) the RLS DAX filter on a table for a role.

Add (or update) a member (user/group) on a role.

member_type: 'external' (default, ExternalModelRoleMember — recommended for Power BI service) 'windows' (WindowsModelRoleMember — domain principal SID)

With overwrite=False (default), raises PowerBIDuplicateError if a member with the same member_name already exists on the role. With overwrite=True the existing member's identity_provider is refreshed (when external) and the response carries action="updated".

Remove a member from a role, matching by MemberName (case-insensitive).

List calculation groups and their items.

Create a calculation group table with its Name column and optional calculation items.

items: list of {name, expression, format_string_expression?, ordinal?}.

Delete a calculation group table.

Add any visual to a report page — the single entry point for visual creation.

visual_type: card, labelled_card, multi_row_card, bar_chart, stacked_bar_chart, stacked_column_chart, clustered_column_chart, hundred_percent_stacked_bar_chart, hundred_percent_stacked_column_chart, ribbon_chart, line_chart, area_chart, stacked_area_chart, hundred_percent_stacked_area_chart, donut, pie_chart, treemap, funnel, table, waterfall, scatter_chart, combo_chart, slicer, gauge, kpi, matrix, map, text_box. config: per-type keys — categorical charts: category_column + value_measure (+ legend_column); axis charts: axis_column + value_measures (list); card/gauge: measure; table: columns (list); matrix: rows + values (lists); scatter: category_column + x_measure + y_measure; combo: category_column + bar_measures + line_measures; kpi: indicator_measure + trend_column; slicer: column (+ slicer_type); map: location (+ measure); text_box: text. Error messages name any missing key. Prefer pbi_add_visual_from_intent when you have a business intent rather than an exact type. dry_run: when True, run all validation and binding logic but skip the layout disk write — response carries dry_run=True and a write_log.

Connect to Power BI Desktop and report the active instance.

List discovered Power BI Desktop instances without forcing a connection.

List model tables and columns.

List all model measures.

List model relationships.

Execute a DAX or DMV query.

Execute a DAX query under a specific role context.

Execute a DAX query and return result rows with timing diagnostics.

Create or update a DAX measure.

Delete a DAX measure.

Create or update a single-column relationship.

With overwrite=False (default), raises PowerBIDuplicateError when a relationship between the same endpoint columns already exists. With overwrite=True the existing relationship is updated in place (cardinality / direction / is_active refreshed) and the response carries action="updated".

Return a full model snapshot in a single call.

Return the most-recent N tool operations recorded by the connection manager (newest first).

Each entry: {ts, op, kind: "read"|"write", duration_ms, ok} plus error_type/error_code/error_message when ok is False. Use this to self-diagnose what just happened — e.g. an LLM can pull the last 5 calls after a failure to see which writes already landed.

Single-call self-diagnostic for stability and dependency status.

Read-only. Skips any TOM/ADOMD probe gracefully when no connection is active. Use as a preflight from any LLM agent: one call answers "can I talk to Power BI right now?" without juggling connect / list / model_info.

Trigger a model or table refresh.

Parse a .dax file and bulk-create measures.

Batch-create or update multiple DAX measures with a single SaveChanges call.

Each item in measures must have at minimum: name, expression. Optional keys: format_string, description, display_folder, is_hidden.

With dry_run=True, every measure is name + expression validated and the per-item planned_action is reported (would_create/would_update/ would_fail), but no model mutation happens — SaveChanges is never called. Use this for an LLM preflight before committing the batch.

Create a family of time-intelligence measures from a base measure.

Default patterns: YTD, MTD, QTD, SPY, YOY, YOY %, MA3. Each generated measure is named "{base_measure} {suffix}" (e.g. "Sales YTD") and lives on table (typically the same fact table as the base measure). Dependency-aware: requesting YOY% automatically adds YOY and SPY so the whole family resolves.

With format_inherit=True, each measure inherits base_measure's format string (best-effort lookup on the live model). format_string overrides that when supplied. Patterns that prescribe their own format (e.g. YOY % → "0.00%") win unless format_string is explicit.

With dry_run=True no model mutation occurs — the response carries a plan listing every measure that would be created/updated/skipped.

Create a period-over-period variance measure.

DAX template (parametrised by granularity ∈ {year, month, quarter}):

[{base}] - CALCULATE([{base}], DATEADD({date_table}[{date_column}], {offset}, {granularity}))

Default compare_period_offset = -1 ⇒ "current period vs previous one".

Create a %-of-total contribution measure.

DAX template:

DIVIDE([{base}], CALCULATE([{base}], ALL({scope_columns})))

scope_columns defines the denominator scope — typically the dimension columns whose total you want each row to be a percentage of (e.g. ["Categorie.Nom catégorie"] for "this category's % of all categories").

Create a Top-N filter measure.

DAX template:

IF(RANKX(ALL({dim_table}[{dim_column}]), [{rank_measure}], , DESC) <= {N}, [{base}], BLANK())

Use as the value of a chart visual to surface only the top N members of a dimension. rank_measure defaults to base_measure.

Create a trailing rolling average measure.

DAX template:

AVERAGEX(DATESINPERIOD({date_table}[{date_column}], LASTDATE({date_table}[{date_column}]), -{window}, {granularity}), [{base}])

Apply a named format preset to a list of measures or columns.

Wraps pbi_set_format_tool so callers don't need to memorise the raw format string. Supported preset families: currency_eur(_k|_m), currency_usd, percent (0dp/1dp/2dp/4dp), thousands, millions, decimal_2, integer, integer_no_sep, date_iso, date_short_fr, date_short_us, date_long_fr, datetime_iso. Use pbi_list_format_presets_tool to inspect them.

Return the catalogue of format-string presets.

Optional filter_substring is matched (case-insensitive) against each preset's name to narrow the result — useful when an LLM only needs the "percent_*" or "currency_*" subset.

Create or update a calculated table.

Create or update a calculated column.

Set the DataType (and optionally FormatString) of an existing column.

Works for any column kind (source, calculated, calculated table column). Use when Power Query type hints (Int64.Type etc.) are overridden by PBI's downstream inference and the column ends up as the wrong type.

data_type accepts standard TOM names (Int64, Decimal, Double, String, DateTime, Boolean, Currency, Binary, Variant).

Batch-apply format strings to measures or columns.

Export the full model as JSON, optionally writing it to disk.

Render a compact markdown snapshot of the model — tables, columns, measures, relationships — ready to paste into an LLM system prompt so the LLM can author DAX with full schema context in one round-trip.

Sections:

• # Model: <name>

• one ## <Table> per table with Columns: and (optionally) DAX Measures: lines

• ## Relationships with A[X] → B[Y] (cardinality, active) rows

Output is truncated to max_chars (default 12 000) with a trailing note when truncation kicks in. Set include_dax=False to omit the measure expressions and stay terse.

Audit the model for common issues: empty expressions, missing format strings, orphan tables, duplicate measure names.

Read the M expression for a specific table partition.

List table partitions and their current source expressions.

Make data sources portable via a Power Query parameter.

Creates (or updates) an M parameter named parameter_name whose default value is file_path. Then rewrites every M partition that string-matches file_path so it calls File.Contents(<parameter_name>) instead of the hardcoded path. After the parameter is wired in, a collaborator can change the file location once via Power BI Desktop's Transform data → Manage parameters dialog without editing each query.

partitions: optional explicit list of "Table[/Partition]" to rewrite. By default every M-source partition that references file_path is rewritten.

Bulk-rewrite a hardcoded file or folder path inside every M partition.

Use when a workbook moves and queries break with DataSource.NotFound. Looks for old_path as a substring of each M-source partition expression and replaces it with new_path. Calculated/query-source partitions are skipped. dry_run=True returns the planned changes without writing.

Write or update an M expression on a table partition.

Generate and inject an Excel import Power Query for a table.

Generate and inject a CSV import Power Query for a table.

Generate and inject a folder import Power Query for a table.

Bulk-create Excel import queries for multiple tables.

Import an Excel workbook into Power BI tables in one call.

List TMDL files in a Power BI Project semantic model definition folder.

Create or replace a measure block in one table TMDL file.

Read one TMDL file from a Power BI Project definition folder.

Create or overwrite one TMDL file inside a Power BI Project definition folder.