krusgraphFor Microsoft Fabric.
Docs menu

How it works

Krusgraph sits between the AI client you ask in and Microsoft Fabric. It reads table and column metadata, makes one plan of keys and relationships with the rule behind each one, counts checks against the rows as queries in Fabric, and creates Fabric items from that plan only when you ask. Four parts, with how far each has been tested:

Clientswhere you askClaudetested end to endVS Codetested end to endCursortested end to endClaude CodeexperimentalGlamaexperimentalLovable, on the APItested end to endMCP, signed in as youKrusgraphone plan: keys and relationships, each with its rule, counted against the rowsPlan and checktested end to end on 27 October 2026creates when you askFabric itemsfrom the same plan, a dry run firstGraphQL APIupdated in place live; anew one: experimentalPower BI semantic modelcreated live; reviewed inFabricFabric IQ ontologycreated live; reviewed inFabricViewsdraft; deployment not yetverifiedData agentdraft; deployment not yetverifiedSilver notebookscreated livebuilt on your tablesOneLakewhere your tables stay; Krusgraph copies none of themLakehousetested end to endWarehouseexperimentalSQL databaseexperimentalreads metadata, counts rows
Clientswhere you askClaudetested end to endVS Codetested end to endCursortested end to endClaude CodeexperimentalGlamaexperimentalLovable, onthe APItested end to endMCP, signed in as youKrusgraphone plan: keys and relationships, eachwith its rule, counted against therowsPlan and checktested end to end on 27 October 2026creates when you askFabric itemsfrom the same plan, a dry run firstGraphQL APIupdated in placelive; a new one:experimentalPower BIsemantic modelcreated live;reviewed inFabricFabric IQontologycreated live;reviewed inFabricViewsdraft; deploymentnot yet verifiedData agentdraft; deploymentnot yet verifiedSilvernotebookscreated livebuilt on your tablesOneLakewhere your tables stay; Krusgraphcopies none of themLakehousetested end to endWarehouseexperimentalSQL databaseexperimentalreads metadata, counts rows
The four layers and how far each has been tested, from the same list as the test status.
  • Clients: where you ask. Claude, VS Code and Cursor connect to Krusgraph through MCP and sign in with your Microsoft work account (VS Code and Cursor at /mcp/connect): tested end to end as clients. Claude Code and Glama are experimental as clients. Lovable builds an app on the GraphQL API Krusgraph creates, not on Krusgraph itself.
  • Krusgraph: the MCP server in the middle. It makes the plan from table and column metadata (keys and relationships, each with the rule behind it), counts the checks as queries in Fabric and writes the drafts; it stores no table contents. Reading, planning and checking: tested end to end on 27 October 2026.
  • Fabric items: what Krusgraph creates in your workspace when you ask, each from the same plan and each with a dry run first. How far creating each one has gone: the API for GraphQL (updated in place live; a new one: experimental), the Power BI semantic model (created live; reviewed in Fabric), the Fabric IQ ontology (created live; reviewed in Fabric), the views (draft; deployment not yet verified), the data agent (draft; deployment not yet verified), the silver notebooks (created live).
  • OneLake: Fabric's storage, where your lakehouse and warehouse tables live (a SQL database in Fabric is replicated there). Krusgraph reads your tables where they are and copies none of them. Lakehouse is tested end to end as a source; Warehouse and SQL database are experimental.

The status words come from the same list as the tool reference, and change there first.