Krusgraph

Reads the plan as a graph: each table's role (fact, dimension, bridge, isolated), the most central tables, subject areas, cycles, table pairs with more than one path, and which relationships can be active together. Requires workspace and source. Returns JSON. Metadata only, no data queries; changes nothing. Errors: {error, kind, hint}.

Turns a question in Swedish or English about given tables into T-SQL over the plan's safest join path, and with api into GraphQL in that API's field names, or says why GraphQL cannot answer it. Requires workspace, source, question and tables. Returns JSON with the query, each join step with its rule, and what it assumed. Runs nothing; query_graphql runs the GraphQL version. Errors: {error, kind, hint}.

Checks what an API for GraphQL item still needs before an app can use it: keys, relationships, write access, introspection, permissions. Only for an API the user names; a schema's check is validate_plan. Requires workspace and an API. Returns JSON {api, endpoint, checks, missing, unknown, verdict}, each with its next step. Changes nothing. Errors: {error, kind, hint}.

Generates a starter pack for an app on an API for GraphQL item: a Lovable prompt, connection details, five GraphQL queries that follow the relationships, a Supabase edge function and the portal steps left. Requires workspace and an API that create_backend has created. Returns JSON {lovable_prompt, connection, queries, checklist, backend, note}. Changes nothing; app_readiness lists what the API lacks. Errors: {error, kind, hint}.

Compares an existing Power BI semantic model with the plan: missing, contradicted and extra relationships, cardinality, direction, active, stale columns, suggested measures. Requires workspace, source and model (SemanticModel.ReadWrite.All or Item.ReadWrite.All) or tmdl (no permission). Returns JSON with the differences and confirmed entries for extra links. Changes nothing. Errors: {error, kind, hint}.

Creates a Fabric API for GraphQL item over the source's tables with the plan's keys and relationships, or updates the API of that name in place. Requires workspace, source and name. Returns JSON {summary, dry_run, result, card (endpoint, portal steps left), skipped_relationships}; the dry run returns instead whether the item exists and would be updated or created, the definition and what changes in it, and the checks that passed. Changes Fabric only with dry_run=false: an update replaces the API's definition; tables outside tables or gone from the source leave it, field names of unchanged relationships are kept, and no item is deleted. When an update would drop relationships the API has, nothing is written and the answer (kind would_remove) lists them as confirmed entries. It does not run validate_plan; plan_backend proposes, validate_plan checks, plan_drift previews an update. Errors: {error, kind, hint}.

Creates a Fabric data agent over the source, with the plan as its instructions (business names, the joins that are safe and those that are not, header amounts not to sum over lines), the plan's tables selected and 3–5 example question/T-SQL pairs over safe joins, or updates the agent of that name in place. Only joins validate=true counted are called safe. Requires workspace, source and name; Item.ReadWrite.All or DataAgent.ReadWrite.All. Returns JSON with the instructions, examples, sizes against the documented limits and the REST calls; with dry_run=false also the result, or kind consent. Changes Fabric only with dry_run=false: instructions, data source and examples are replaced; nothing is deleted. Errors: {error, kind, hint}.

Creates drift_monitor_draft's notebook and pipeline, or updates the items of those names in place; with schedule=true also adds a daily schedule unless one at that time exists. Requires workspace, a lakehouse source and name; Notebook.ReadWrite.All and DataPipeline.ReadWrite.All (or Item.ReadWrite.All), and Contributor. schedule is false by default: the schedule needs DataPipeline.Execute.All, opt-in on the hosted app. Returns JSON with the items and REST calls; with dry_run=false also each step's result. Changes Fabric only with dry_run=false: the two items' definitions are replaced, existing schedules are kept, nothing is deleted and no secret is written. Errors: {error, kind, hint}.

Creates a Fabric notebook item with gold_views_draft's notebook, or updates the notebook of that name in place. Requires workspace, a lakehouse source with schemas and name; Notebook.ReadWrite.All or Item.ReadWrite.All and Contributor. Returns JSON with the views, the notebook and the REST call; with dry_run=false also the result. Changes Fabric only with dry_run=false: the notebook's content is replaced, no item is deleted, and nothing runs; when run, it stops before writing over a table it did not create. Errors: {error, kind, hint}.

Creates a Fabric IQ ontology (preview) item from the plan, or updates the item of that name in place. Requires workspace, source and name; the delegated permission Item.ReadWrite.All, Contributor on the workspace and a Fabric capacity. Returns JSON with the definition parts and the REST call; with dry_run=false also the result, or kind consent when the permission is missing. Changes Fabric only with dry_run=false: an update replaces the item's definition, so entity types not in the plan leave it; no item is deleted. Errors: {error, kind, hint}.

Creates the pipeline export_item_definitions drafts with items=["pipeline"]: the silver and gold notebooks of create_silver_notebook and create_gold_views, a check notebook, and a data pipeline that runs them in that order, each only after the one before succeeded, so a failed check fails the run. Items of these names are updated in place. Requires workspace, a lakehouse source with schemas and name; Notebook.ReadWrite.All and DataPipeline.ReadWrite.All (or Item.ReadWrite.All), and Contributor. Returns JSON with the definitions and REST calls; with dry_run=false also each step's result. Changes Fabric only with dry_run=false: four definitions are replaced, nothing is deleted; nothing runs and no schedule is added unless run_once or schedule asks (both off by default; they need DataPipeline.Execute.All, and confirm_jobs from the dry run after the person said yes). Errors: {error, kind, hint}.

Creates a Power BI report item bound to an existing semantic model: one page with cards, a column chart by month, a bar chart and a table built only on the model's measures and columns (PBIR definition), or one blank page with empty_page=true. Requires workspace, model and name; Report.ReadWrite.All or Item.ReadWrite.All, Contributor, and to read the model SemanticModel.ReadWrite.All. Returns JSON with the visuals, the REST call and next steps; with dry_run=false also the result, or kind consent. Changes Fabric only with dry_run=false, and only by adding one item: a report of that name is left as it is (kind already_exists), nothing is updated or deleted. Errors: {error, kind, hint}.

Creates a Fabric semantic model item from semantic_model_draft's TMDL (Direct Lake), or updates the item of that name in place. Requires workspace, source and name; SemanticModel.ReadWrite.All or Item.ReadWrite.All and Contributor on the workspace. Returns JSON with the parts and the REST call; with dry_run=false also the result, or kind consent. Changes Fabric only with dry_run=false: an update replaces the model's definition, so tables, relationships and measures added outside the plan leave it; no item is deleted. Errors: {error, kind, hint}.

Creates a Fabric notebook item with silver_notebook_draft's notebook, attached to the lakehouse, or updates the notebook of that name in place. Requires workspace, a lakehouse source and name; Notebook.ReadWrite.All or Item.ReadWrite.All and Contributor. Returns JSON with the notebook and the REST call; with dry_run=false also the result, or kind consent. Changes Fabric only with dry_run=false: the notebook's content is replaced, no item is deleted, and nothing runs; when run, it writes only in silver_schema. Errors: {error, kind, hint}.

Creates or alters wide T-SQL views over the plan on the source's SQL analytics endpoint or warehouse: one per fact table with its dimensions, one per header table. Requires workspace and source; rights to create objects (Contributor is enough). Returns JSON with the SQL; with dry_run=false also what ran. Changes Fabric only with dry_run=false and only in views_schema: CREATE OR ALTER replaces views of the same name; nothing is dropped. Errors: {error, kind, hint}.

Writes hand-over documentation: tables, columns and types, keys and relationships with their rules, keyless tables, ambiguous columns, the ER diagram and an empty Description column. Requires workspace and source. Returns Markdown text, cut on a line above 60 000 characters. Changes nothing. Errors: {error, kind, hint}.

Writes a dbt sources file: unique and not_null on key columns, relationships tests on single-column relationships, composite ones as comments. project=true: a dbt project draft for dbt-fabric 1.12 on dbt-core 1.12 (Fabric Warehouse T-SQL): typed, de-duplicated staging models; marts per domain joined only over relationships check_rows counted many-to-one, child facts summed before the join; untemplated parts and TODOs listed. quality: the same checks as Great Expectations and/or Soda drafts; profile: a Fabric profiles.yml. Can sit beside dbt Labs' dbt-mcp, which runs dbt. Requires workspace and source. Returns YAML text, or JSON files for a project. Read-only (check_rows counts rows); creates nothing, runs no dbt. Errors: {error, kind, hint}.

Describes an API for GraphQL item in the names a query uses: types, list query fields, fields and relationships. Requires workspace and an API that create_backend has created (list_sources lists them). Returns JSON {types: [{type, query_field, source, fields, relationships, by_key, writable}], how_to_query}. Read from the item's definition, so introspection can be off; changes nothing. query_graphql runs a query. Errors: {error, kind, hint}.

Proposes domains from the plan. Returns JSON with interfaces and an ODCS contract per domain. Changes nothing. Errors: {error, kind, hint}.

Drafts a nightly Fabric notebook and pipeline that read the lakehouse schema, compare the plan with the last one saved in krusgraph.plans and post a summary to a webhook set as a pipeline parameter. Requires workspace and a lakehouse source. Returns JSON with the notebook, the pipeline definition and the schedule. Changes nothing; create_drift_monitor creates the items. Errors: {error, kind, hint}.

Exports what the creating tools would create as files in Fabric's Git format: a . folder per item, views as .sql, a README; items=["pipeline"] adds a data pipeline draft (silver, gold, then a check that fails the run); items=["catalog"] the OneLake catalog descriptions and tags per item with the REST calls (not sent). Requires workspace and source. Returns JSON with the files, the items and what was skipped and why. Changes nothing. Errors: {error, kind, hint}.

Drafts materialized lake views (Spark SQL) over the silver tables: one wide view per fact table in gold_schema, dimensions joined over many-to-one paths, the plan's keys and relationships as CHECK rules. Requires workspace and a lakehouse with schemas. Returns JSON with the SQL per view and the notebook that creates them. Changes nothing; create_gold_views creates the notebook. Errors: {error, kind, hint}.

Finds the shortest chain of relationships between two tables. Requires workspace, source, start and end. Returns JSON with each hop and its rule, every equally short path and a T-SQL SELECT … JOIN; without a path, the missing relationship as a confirmed entry. Metadata only; changes nothing. answer_query_draft drafts a query for a whole question. Errors: {error, kind, hint}.

Shows which Fabric objects in the workspace build on a source's tables: GraphQL APIs, semantic models, views on the SQL endpoint and data agents. Requires workspace and source. Returns JSON: a graph (nodes, edges), objects per table, tables nothing uses, items that could not be read, and with table, what a change to it reaches; format="openlineage" returns OpenLineage events instead; runs= returns its latest runs, per-activity status and Fabric's error text with a heuristic category and next step. Read-only: starts, cancels and fixes nothing. Errors: {error, kind, hint}.

Lists the data sources in one workspace (lakehouses, warehouses, SQL databases, mirrored databases and catalogs) and its API for GraphQL items. Requires a workspace. Returns JSON {workspace, sources: [{name, id, type}], graphql_apis: [{name, id}]}; a source's name is the source, an API's name the api of the other tools. Read-only. Errors: {error, kind, hint}.

Lists the tables and columns with types in one schema of a source, and every schema of the source (schemas). Requires workspace and source. Returns JSON (a column count per table in summary form). Read-only. Errors: {error, kind, hint}.

Lists the Fabric workspaces the signed-in account can see. Takes no arguments. Returns JSON {workspaces: [{id, name, capacity}]}; the name or id is the workspace of the other tools. Read-only. Errors: {error, kind, hint}.

Drafts a Fabric IQ ontology (preview) from the plan: entity types with keys, relationship types and bindings. Requires workspace and source. Returns JSON with a summary (the rule behind each key and link), the binding checklist, a Turtle file for Import ontology and the item definition parts. Changes nothing; create_ontology creates the item. Errors: {error, kind, hint}.

Infers primary keys and relationships for the tables in a schema and proposes them as a plan. Requires workspace and source. Returns JSON: summary, keys and relationships each with the rule behind it (K0–K5, R0–R5), ambiguous columns, keyless tables, suggested composite keys and a Mermaid ER diagram. Rule R5 counts code and role columns the names left unlinked against the rows (read-only counting queries, at most 20, over the first 200 distinct values of a column); candidates it could not confirm are listed as proposed. Changes nothing in Fabric. validate_plan checks it against the rows; create_backend builds the API. Without schema, a source with several schemas that hold tables gets {choose_schema, schemas: [{schema, tables}], summary, hint} and nothing is planned (n169). Errors: {error, kind, hint}.

Compares today's plan with an existing API for GraphQL item: new tables and columns, mapped columns the source no longer has, relationships the plan no longer supports, and whether create_backend with that name would change anything. Requires workspace, source and api. Returns JSON; stops reading what builds on the tables after 40 s and lists what it did not read (time_limit). Changes nothing. Errors: {error, kind, hint}.

Runs a read-only GraphQL query against an API for GraphQL item; given question instead of query, it builds the query as answer_query_draft does and runs it. Requires workspace and api, plus query, or question with tables and source. Returns JSON with the data, size-bounded, and for a question the query used. Mutations and subscriptions are refused; changes nothing. describe_backend gives the field names. Errors: {error, kind, hint}.

Generates Rayfin (Fabric Apps) connector entity files and schema.ts from the plan, with keys and @one/@many relationships. Requires workspace and source. Returns JSON {path: file content}. Changes nothing in Fabric. Errors: {error, kind, hint}.

Builds a review file of the plan: each key and relationship with a review status and a check status, plus a Markdown report. Requires workspace and source. Returns JSON with the file, the report and what needs attention first, no row values. Changes nothing. Errors: {error, kind, hint}.

Returns the first rows of one table so that a column's meaning can be checked. Requires workspace, source and table. Returns JSON {table, columns, rows, personal_data}; long text is shortened, and columns that look like personal data are masked unless show_personal_data=true. Read-only. Errors: {error, kind, hint}.

Shows which tables of the source the SAP pack recognises (VBAK, VBAP, LIKP, VBRK, KNA1…, with or without a prefix such as sap_), their keys with MANDT, the order-to-cash relationships it applies with the public source of each, the ones it lists but does not apply (VBFA), missing pack tables and columns, and the traps (client, header vs item amounts, currency, leading zeros, partner functions). Requires workspace and source. Returns JSON, with the chosen export when export is not "none". Changes nothing. Errors: {error, kind, hint}.

Drafts the plan as a Power BI semantic model in TMDL (Direct Lake on the source's SQL analytics endpoint): relationships with cardinality and filter direction, one active path per table pair, suggested measures that sum at the right level (validate=true also counts header amounts repeated in a flat table). Requires workspace and source. Returns JSON {path: content} for the definition folder plus README.generated.md listing the decisions to check. Changes nothing; create_semantic_model creates the item, compare_semantic_model checks an existing model. Errors: {error, kind, hint}.

Drafts a PySpark notebook writing silver Delta tables: one row per key unless the key is only guessed, references checked, broken rows to krusgraph_quarantine. Requires workspace and a lakehouse source. Returns JSON with the steps and the notebook source. Changes nothing. Errors: {error, kind, hint}.

A guided path for a first or vague message ('start', 'start Krusgraph', 'help', 'what can you do', 'check my data'): workspace, source, schema, map, check, decide, doubled totals, a dry run of all builds with the data flow, then create after a second confirmation of the names. Returns JSON {step, path, actions: [{tool, arguments, say}], after}: each action is an existing tool; after is the next start call. Changes nothing. Errors: {error, kind, hint}.

Finds columns that look like references to another table's key but are in no relationship, ranks them and counts the best against the data (share of values found, distinct values). Requires workspace and source. Returns JSON candidates, each with a confirmed entry for plan_backend or create_backend. Read-only counting queries; applies nothing. Errors: {error, kind, hint}.

Checks the plan against the rows: is each key unique, non-null, how many references find no target (orphans). For "check shop", "run a check". Requires workspace and source; keys, confirmed as in plan_backend. Returns counts and each item's evidence state, never row values; past the call's time budget, "not checked (time)" and a resume_from. Read-only. Errors: {error, kind, hint}.