Skip to main content
Læs siden på dansk The Power BI connector gives your AI read-only access to Microsoft Power BI. Your AI can find your workspaces, reports and dashboards, read a dataset’s tables, columns and measures, and fetch the numbers straight from the dataset with a DAX query. Then it can answer questions like “what was revenue per region in Q3” with the numbers your reports are built on. You connect with your Microsoft work account, and your AI never sees the access itself.

Access

Read-only, read permissions only

Authentication

Microsoft work account

Hosting

Hosted in the EU · Frankfurt
  • Price: 299 DKK/month excl. VAT (€39/month if you have chosen the EU · EUR market in the portal). No lock-in, cancel at any time. The price covers 1 user and 1 agreement (one connected Microsoft user). Each extra user and each extra agreement costs 149.50 DKK/month excl. VAT (€19.50). Also included in the Consile Studio membership and the free edition. See Pricing.
  • Users: whoever buys Power BI automatically gets the included access. If more colleagues need Power BI in their AI, an administrator gives them access under Team in the portal (choose All tools next to Power BI on the person’s page). An extra user is billed pro rata from the day access is given. A colleague without access can still see the tools in their AI, but every lookup is refused with a message to ask an administrator for access. If you have the Consile Studio membership, an administrator instead gives one membership licence per person, and it also covers Power BI. See Team.
  • Multiple Microsoft users and tenants: the portal calls each connection to the system an agreement. One agreement is one connected Microsoft user and covers the workspaces that user has access to. You need one more agreement if your AI should also see things with another user’s access, or if you have Power BI in several Microsoft tenants. An administrator clicks Add agreement on the Connections card on the connector’s page in the portal. Each agreement gets its own address, for example https://mcp.consile.ai/power-bi-2/mcp, which you add as a separate connector in your AI. Under the Consile Studio membership and in the free edition, other rules apply to adding an agreement. See Connect Power BI.
  • Endpoint: https://mcp.consile.ai/power-bi/mcp (your own named connector in your AI).

​
What does it require?

  • Power BI in Microsoft’s cloud. The connector reads from the Power BI service (app.powerbi.com) through the Power BI REST API.
  • A Microsoft work or school account. You log in with an account in your organization at Microsoft. A personal Microsoft account cannot be used.
  • A Power BI user with access. Your AI only sees the workspaces, reports and datasets that the connected user can see in Power BI. For your AI to run DAX queries against a dataset, the user needs Build permission on that dataset.
  • Often an administrator’s approval. In many organizations, Microsoft requires an administrator to approve the Power BI permissions. See Connect Power BI.

​
Prerequisites

1

Subscription

The connector must be bought in the portal. See Buy a connector.
2

Microsoft approved

Log in with your Microsoft work account, and approve Consile at Microsoft. The connection applies to your whole Consile account. See Connect Power BI.
3

Access for your user

If there are several users on the account, an administrator must assign you Power BI under Team in the portal. Whoever bought the connector has access from the start. If you have created your own Consile account, but belong to a company that already has Power BI, ask your administrator for an invitation to the team instead of buying again. See Team.
4

Your AI connected

Each user adds the connector in their own AI and logs in with their own Consile login. See Connect Claude or Connect ChatGPT.

​
Read-only

All the permissions you approve at Microsoft are pure read permissions: Workspace.Read.All, Report.Read.All, Dataset.Read.All, Dashboard.Read.All, App.Read.All and Dataflow.Read.All, plus offline_access, which keeps the connection alive. In addition:
  • The connector has no write tools. All twelve tools can only read.
  • Consile only sends read requests to Power BI.
  • DAX queries can only read. A query must start with EVALUATE (optionally with DEFINE in front), and it cannot change a dataset.
  • If Microsoft’s approval contains permissions for anything other than Power BI, the connection is refused.
Consile does not store data from Power BI. Each answer is fetched directly when your AI asks.

​
Workspaces and “My workspace”

Power BI groups reports, dashboards and datasets in workspaces. Your AI finds them with list_workspaces, which shows the workspaces the connected user is a member of. The user’s personal “My workspace” is not on the list. Your AI reads it by not naming a workspace. The numbers live in the dataset, not in the report. A report is a layout on top of a dataset. So your AI typically finds the report or the dataset first, reads the dataset’s schema with get_dataset_schema, and then fetches the numbers with execute_dax_query.

​
What can it do? (tools)

You do not need to know the tool names. You ask in natural language, and your AI picks the right one itself. The connector provides twelve read-only tools. When your AI calls them, they are called power-bi__<tool> (for example power-bi__list_workspaces). Workspaces and apps Reports and dashboards Datasets

Building on top yourself? See the API reference

Full reference with the twelve tools, arguments, result fields, DAX rules and error messages.

​
Try asking

“What was revenue per region in Q3 according to the Sales dataset, and how does it compare with Q2?”
“Find the ten customers with the highest revenue this year in the Sales dataset.”
“Which reports exist in the Management workspace, and which dataset is each of them built on?”
“Which tiles are on our sales dashboard, and where do the numbers come from?”
“When were the datasets in the Management workspace last refreshed, and are any refreshes failing?”
“Where does the Sales dataset get its data from?”
Mention the workspace and the name of the dataset or report in your question. Ask for totals and top lists rather than all rows. Then the answer is faster, and your AI gets all of it.

​
Limitations

  • Read-only: the connector cannot create, change, delete or refresh anything in Power BI. It cannot start a dataset refresh either.
  • Work accounts only: personal Microsoft accounts are not supported.
  • The user’s access applies: your AI only sees the workspaces the connected user is a member of, and only the apps the user has installed. If the user lacks access to a workspace or Build permission on a dataset, Power BI answers with an error (403), and your AI is told about it. Reconnecting does not help. Instead, ask the owner of the workspace for access.
  • No visuals: the connector does not show report pages or tiles as images. Your AI reads the numbers from the dataset with a DAX query.
  • Large answers are cut off: execute_dax_query returns 200 rows by default and at most 2,000. If there are more, the rest are cut off, and the answer says so. Ask for totals or a top list instead. A DAX query may be at most 20,000 characters.
  • The schema needs a current dataset: get_dataset_schema does not work on push datasets or very old datasets. Your AI is told so, and a hand-written DAX query still works. Long DAX expressions for measures only show the first 500 characters.
  • Dataflows only in workspaces: list_dataflows needs a workspace. Dataflows do not exist in “My workspace”.
  • Apps and dataflows can be turned off: if your organization has turned off apps or dataflows in Power BI, list_apps or list_dataflows answers with an error (403). The other tools work as normal.
  • Refresh history: a dataset that cannot be refreshed (for example DirectQuery or a live connection) has no history.
  • Microsoft’s call limit: Power BI limits how many calls can be sent. If a lookup hits the limit, your AI is told to wait a little and ask for fewer rows.
  • Explanatory errors, but not always: for the common errors, the connector gives your AI an explanation: an access that must be connected again, missing access, an id that does not exist, and a query that Power BI refuses, for example a DAX error. Other errors, for example a timeout or an error at Microsoft, only give the general error “Tool invocation failed.”
  • The access is renewed automatically, but can stop: if the renewal fails, or Consile is removed in your Microsoft organization, your AI is told that the connection must be set up again from the Consile portal. The portal can still show Connected, because the stored access is still there. Only an administrator can reconnect: if the agreement says Access revoked, click Reconnect. Otherwise the administrator chooses Disconnect in the menu next to the agreement and then clicks Connect. You can also ask your AI for a new connection link in the connector that belongs to the agreement (for example power-bi-2). See Connect Power BI.
Is something not working as expected? See Troubleshooting.