Imagine a sales manager with no DAX knowledge and no BI background, typing “What were our worst-performing regions last quarter?” into a chat window and getting a precise, data-grounded answer in seconds. No ticket to the analytics team. No waiting. Just an answer.
That’s not a future-state demo. That’s what MCP Servers in Power BI make possible today, and in my view, it’s the most significant shift in how organisations interact with their BI data in years.
So, What Exactly Is an MCP Server?
MCP (Model Context Protocol) is an open standard, originally from Anthropic and now widely adopted, that lets AI agents securely connect to external tools and data sources. Think of it as a universal plug between your AI assistant and the systems it needs to work with.
Power BI now ships with two flavours of MCP server:
- Modeling MCP Server (local): Runs on your machine. Let AI agents create, edit, and optimise your semantic model, writing measures, renaming columns, validating DAX, all through natural language.
- Remote MCP Server (cloud): A hosted endpoint that lets AI agents query your published models and surface insights conversationally, using the same Copilot-powered DAX engine you’d find in Power BI service.
The Use Case That Convinced Me
I’ve seen a lot of “AI + BI” demos that fall apart the moment you go beyond a toy dataset. What struck me about MCP servers is a specific scenario: automated model optimisation.
A developer connects Claude Desktop to the Modeling MCP Server, points at a slow-running measure, and says “optimise this.” The agent runs performance traces, proposes changes, applies them, and validates the result, all without the developer touching the model directly. Query times dropped. Capacity usage fell. And critically, the developer could see exactly what changed and why.
That’s not AI as a gimmick. That’s AI doing the tedious, high stakes work that slows down every BI team.
The Trade-offs
I’d be doing you a disservice if I only sold the upside. A few things worth knowing:
- Still in preview: Several features are not yet GA. Plan for change.
- Model quality matters more now: Garbage in, garbage out applies harder when AI is generating the DAX. A messy model produces confidently wrong answers.
- Security needs thought: MCP servers respect Power BI’s RBAC and Azure AD, but integrating third-party LLMs means data can travel outside your compliance boundary. Know your data before you connect it.
- RLS has a gap: Row-level security works for user-authenticated queries, but service principal authentication currently bypasses it. Worth flagging to your security team early.
My Take: Where to Start
If I were introducing MCP servers to my team today, I’d resist the urge to go broad. Pick one high-value, low-risk model, something internal and not customer-facing, and pilot the Remote MCP Server for conversational querying. Let a few analysts use it for a few weeks. The feedback loop from real usage is worth more than any proof-of-concept demo.
Only after that would I explore the Modeling MCP Server for agentic tasks. The potential there is enormous, but it also involves write access to your models, so governance and testing need to come first.
What to Do Next
- Try it: Set up the Remote MCP Server in VS Code with GitHub Copilot Chat. Microsoft’s documentation on Microsoft Learn walks you through it in under an hour.
- Explore the open-source community: Projects like powerbi-mcp and pbi-desktop-mcp-public on GitHub offer reference implementations worth studying.
- Talk to your security team now: Don’t let governance be an afterthought. Bring them in early and you’ll move faster later.
- Start a conversation: I’d love to hear how others are using or thinking about MCP in their BI environments. Drop a comment below or connect with me on LinkedIn.
The way we interact with BI data is changing faster than most people realise. MCP servers are a big part of that story, and the best time to start understanding them is now.
How Coeo Can Help
If you’d like to know more about how MCP Servers and Power BI can benefit your organisation, reach out at info@coeo.com