MCP connectorsPower BI

Power BI MCP: connect Power BI to Claude and ChatGPT

Ask Claude or ChatGPT what sales were last month against budget, which region is behind or what the margin was on a product line. The figure comes from the same Power BI data your reports are built on, without opening the report and clicking through its filters.

We build the connector for your company on Microsoft’s own Power BI API. Each person signs in with their own Microsoft work account, so the AI sees the workspaces and figures that person may see in Power BI, and it only reads.

Built for your company · Works in Claude, ChatGPT and Copilot

In practice

What your team can do with Power BI in Claude and ChatGPT

Managers

The figure before the meeting

This month against last month and against target, in two sentences you can take into the room. No hunting for the right report page first.
Finance

Actuals against budget, asked your way

Revenue, cost and margin by month, cost centre or business unit, using the measures your reports already use, so the number matches the dashboard.
Sales leads

Who is ahead and who needs help

Sales by rep, region or customer for the period you name, compared with the same period last year.
BI teams

Fewer “can you pull this for me” requests

People find the right report and get the simple figures themselves. You keep one definition of each measure, and every question is logged.
Examples

Ask it like this

  • “What were total sales in October against budget, and how does that compare with September?”

  • “Which three regions are furthest behind their target this quarter?”

  • “What was gross margin per product line last month?”

  • “Which report shows customer churn, and which workspace is it in?”

  • “How much has Pembroke Retail bought from us this year compared with last year?”

  • “When was the Sales Overview data last refreshed, and did the refresh succeed?”

The connector

What we build for Power BI

Deployed OS has no ready Power BI connector yet. We build it for your company when you start, on Power BI’s own API, with the same rules as every other connector: each person’s own login, access decided by group, every call logged.

Workspaces and reports
The workspaces, reports and dashboards you have access to, with names, owners and links, so the AI can point you to the right one.
Semantic models
The data models behind your reports: their tables, columns and the measures your BI team has defined.
The figures behind a report
Read-only queries against a semantic model, so a question is answered with the same measures the report uses.
Refresh status
When a model was last refreshed and whether the refresh succeeded, so you know how fresh the figure is.
Your company’s terms
What you call things, such as net sales, order intake or contribution margin, mapped to the right measure.
Security

Who sees what

Read-only
The connector reads. It does not change reports, edit models or start refreshes. We only build write actions if you ask for them.
Each person’s own Microsoft account
Everyone signs in with their own work account. Power BI applies that person’s workspace access and the row-level security on the model, so a sales rep who only sees their own region in the report sees the same through the AI.
Your company decides who gets it
Admins choose which groups get Power BI in Deployed OS. A group without access never sees the connector.
Every call logged
Who asked, which model was read and when. Deployed OS keeps the log, and it can be exported.
Setup

Set up the Power BI connector in three steps

01
We go through with you which reports and figures people ask about most, and build the connector on the Power BI API.
02
Your Microsoft administrator approves the connection for your company, and you choose which groups get Power BI in Deployed OS.
03
Each person connects with their own Microsoft work account. From then on they just ask.
Compared

Power BI’s own MCP server or Deployed OS?

Microsoft has its own MCP servers for Power BI. The remote one is a hosted endpoint, in preview, that lets AI agents query your existing semantic models with a Microsoft sign-in once a Power BI admin has enabled it. A second, local server is for the people who build and change models. If Power BI is the only system your team wants in its AI tool, Microsoft’s server is a good place to start.

Deployed OS is for when Power BI is one system among several. It sits next to your CRM, finance system and other tools in one company connection, admins decide which groups get each system, every call is logged in one place, and the same rules apply in Claude and ChatGPT. We also build the tools around your own measures and terms, so people can ask in the words they use.

Their documentation: Power BI MCP servers

Questions

Power BI MCP: common questions

Yes. Microsoft offers its own Power BI MCP servers, including a hosted one in preview for asking questions of existing semantic models. Deployed OS adds Power BI to one company connection with your other systems, with access per group, one log and tools built for your company.

No. The connector reads workspaces, reports, semantic models and the figures in them. It does not edit or publish anything, and we only build write actions if you ask for them.

Yes. Everyone signs in with their own Microsoft work account, and Power BI checks each call against that person’s workspace access and the row-level security on the model. The AI cannot see more than the person could in Power BI.

It reads the data model behind the report, not a picture of the page. That is why the answer uses the same measures as the report, and why you can ask for a breakdown the report does not show.

Deployed is an EU company and fully GDPR compliant, and you choose where your data is stored, down to the city. Figures from Power BI are fetched when someone asks for them.

It depends on the size of your team. Get in touch and we will walk you through it.

More connectorsAll connectors
Let’s talk

See it on your own tools.

Tell us which systems your company runs on. We’ll show you what your team can ask, and what IT gets to control.