Skip to content

Use BI Pixie in Claude

With the BI Pixie connector, Claude works with your Power BI data as you. Claude can find and read the semantic models you can open, ask BI Pixie which fields and values people use in the reports built on them, and run queries as you, within your own permissions and row-level security. Claude can also read your BI Pixie account, AI Readiness results and benchmarks, and start a benchmark of Claude.

What BI Pixie adds to Claude

On the AI Assistants page, the Use BI Pixie in Claude card lists four uses. This page explains each of the four.

The Use BI Pixie in Claude card on the AI Assistants page. One sentence says Claude answers questions about your semantic models the way your reports do. Four uses follow, two by two: ask questions about your data, answer the way your reports do, test how well Claude answers, and manage BI Pixie from Claude. Below them, the connector address for the account has a Copy address button. The card ends with an Open Benchmarks button and a link named Learn more about BI Pixie in Claude.

Ask questions about your data

Claude finds your semantic models and answers questions about them, signed in as you, so row-level security applies.

Answer the way your reports do

BI Pixie tells Claude which fields and values people use in the reports built on each semantic model, and gives Claude the semantic model's descriptions. For people who may edit the semantic model, Claude also receives its AI instructions and AI data schema, and BI Pixie matches the synonyms written for it. See Considerations and limitations.

Test how well Claude answers

You can benchmark Claude on a semantic model, in BI Pixie or in the Claude app, and compare runs in an A/B test. On the card, Open Benchmarks opens the Benchmarks tab of AI Readiness. See Benchmark Claude.

Manage BI Pixie from Claude

With the BI Pixie MCP server running on your own computer, Claude Desktop and Claude Code can assess semantic models, Add Pixies and set schedules for you. BI Pixie shows you each change and asks you before it acts. See MCP Server Setup.

Prerequisites

  • A BI Pixie account, on any plan.
  • A Claude plan that allows custom connectors. On a Claude Team or Enterprise plan, an owner of your Claude organization may have to add the connector first.
  • Read or Build permission on the semantic models you ask about.
  • For answers that follow how your organization reads its data, Pixies on the reports built on those semantic models, and the BI Pixie Dashboard installed in a Fabric workspace.

Add the BI Pixie connector in Claude

  1. In Claude, open Settings, then Connectors, and select Add custom connector.
  2. Name it BI Pixie, and enter the address: https://mcp.bipixie.com/mcp for an account in the United States, or https://mcp-eu.bipixie.com/mcp for an account in Europe. On the AI Assistants page in BI Pixie, you see the address for your account.
  3. Select Connect, and sign in with your Microsoft work account.
  4. Approve BI Pixie's access. Microsoft's approval screen names the app BI Pixie Claude connector. Microsoft returns you to Claude, where the connector appears as connected.

To use Claude Desktop or Claude Code on your own computer, you can run the BI Pixie MCP server locally instead. See MCP Server Setup.

Ask Claude about your data

  • Ask about the data in a semantic model, such as last quarter's sales in Europe. Claude finds the semantic model, reads it and runs a query as you.
  • Ask which fields and values your organization uses for a term, so that Claude reads "revenue" or "region" the way your reports do.
  • Ask how your BI Pixie account, AI Readiness assessments and benchmarks stand.
  • Ask Claude to benchmark itself by answering the questions in the conversation. See Benchmark Claude.

Before Claude does anything that uses an allowance, such as starting a benchmark, Claude shows you what it will do, and it acts only after you agree in Claude. A benchmark you started in BI Pixie and pasted into Claude needs no second agreement, because you agreed in BI Pixie when you started it.

Considerations and limitations

Read permission and Fabric IQ

  • With Build permission on a semantic model, BI Pixie runs Claude's queries directly, as you.
  • With Read permission only, BI Pixie answers through Microsoft's Fabric IQ, still as you and with your row-level security. The first time that happens, BI Pixie asks you to grant two Fabric permissions, by name.
  • If your organization requires an administrator's approval for those permissions, Claude tells you what is missing. Build permission on the semantic model lets BI Pixie answer without them.
  • Where Fabric IQ is not offered, such as in a sovereign cloud, Claude needs Build permission, and the answer says so.

The daily limit on the Free plan

  • BI Pixie's cloud is the BI Pixie connector you add in Claude's Connectors settings. Its queries are counted in every Claude app that connects to it, not only on claude.ai. The BI Pixie MCP server you run on your own computer does not use BI Pixie's cloud.
  • On the Free plan, Claude can run 25 queries a day through BI Pixie's cloud, shared by everyone in the account. The count starts again at midnight UTC.
  • On the Usage tab of the Account page, under Claude queries today, you see the count. See Account Settings.
  • Queries from the BI Pixie MCP server on your own computer are never counted. Paid plans have no daily limit.

What an administrator can turn off

  • When an administrator turns off Allow the BI Pixie API and connectors, BI Pixie refuses calls from scripts and from connectors that act for people, for everyone in the account. Usage questions to the local BI Pixie MCP server keep working. See API and connectors.
  • When an administrator turns off Allow Claude to query through BI Pixie's cloud, BI Pixie stops only the queries that Claude runs through BI Pixie's cloud. Claude still reads semantic models and report usage, and the BI Pixie MCP server on your own computer still runs queries. See Claude queries through BI Pixie's cloud.

AI instructions

  • Claude receives the AI instructions written for a semantic model only when you may edit that semantic model. Microsoft gives the definition that holds them only to the people who may edit it.
  • The first time Claude reads AI instructions, BI Pixie asks you to grant one Microsoft permission, by name, to read semantic model definitions. Microsoft names it SemanticModel.ReadWrite.All, because Microsoft gives a semantic model's definition only under a permission that could also change it. BI Pixie uses it only to read the definition and never changes the semantic model.
  • With Read or Build permission only, Claude is told "AI instructions are not available for this semantic model." and answers without them.
  • When Claude reads the AI instructions, BI Pixie also tells Claude which fields were left out of the AI data schema, and for the next 15 minutes matches your terms to the synonyms written for that semantic model. Claude never sees the synonyms themselves.
  • In your Claude app's connector settings, you can turn off the tool named get_ai_instructions. Claude then answers without AI instructions.

What Claude sees

  • The rows your queries return go to Anthropic, because Claude reads them to answer you. BI Pixie does not store those rows.
  • Before Claude reads the rows a query returns, BI Pixie replaces each value that looks like an email address with "(email address withheld)", with a note saying how many were withheld. The rest of each row, and the names of tables, columns and measures, reach Claude as they are.
  • For what BI Pixie sends to Anthropic, see the Privacy Policy.

Related content