MCP Engine Wiki (Power BI Semantic Models)

This wiki explains what the MCP server can do with Power BI semantic models, how to work with it effectively through an LLM, and what’s intentionally out of scope.

This documentation is written for developers, analysts, and consultants who use an MCP client (for example, Claude Desktop) to:

• connect to a semantic model (Desktop PBIX or Service dataset),
• explore and understand it quickly,
• run DAX queries safely,
• make controlled model changes (when allowed),
• validate, test, and govern changes.

How to use this wiki

• If you’re new: start with the workflow pages.
• If you want the core concepts (modes, Desktop vs Service, safe sharing): start with Concepts.
• If you want role-specific guidance: use Choose your path.
• If you want copy/paste prompts: jump to the Prompting playbook.
• If you need exact capabilities per tool: use the tool reference pages.

What “using MCP” looks like (user perspective)

You describe what you want in natural language. The assistant uses tools behind the scenes to:

• connect to the right model,
• browse/search objects,
• run validation queries,
• (when allowed) apply changes safely.

In supported clients (for example, Claude Desktop), some tools can also present a UI (connection browser, query runner, preferences UI).

The "golden" usage pattern (recommended)

1. Connect intentionally (don't auto-select): tell the LLM which model/dataset to use.
2. Explore before acting: have the LLM list/search relevant objects and summarize current state.
3. Plan changes: ask for a brief plan and impact (dependencies) before edits.
4. Apply changes in small batches: prefer one domain at a time (schema vs measures vs security).
5. Validate: re-run key queries, refresh if needed, and (optionally) run unit tests.

First 10 minutes (quickstart)

1. 1

   Connect intentionally

   “List available models/datasets and ask me which one to connect to (don’t auto-select).”
2. 2

   Confirm context

   “Show what we’re connected to (Desktop vs Service, model name/id) and confirm it matches ‘X’.”
3. 3

   Get oriented

   “Summarize the model: main tables, relationships, key measures, calc groups, and roles.”
4. 4

   Find what you care about

   “Search for ‘margin’ across names/descriptions; then expand to expressions if needed.”
5. 5

   Validate with a small query

   “Run a validation query for ‘[Total Sales]’ by month (last 12 months). Return only aggregates.”

If you're in a locked-down environment

Then use these patterns (and see Modes and restrictions):

• Browse-only: focus on schema browsing and documentation; ask for queries you can run elsewhere.
• Read-only: focus on exploration + validation queries + performance analysis; ask for an edit plan you can apply manually if needed.

Quick prompts (start here)

For more templates, see the Prompting playbook.

Licensing and tiers

MCP Engine works on a Free tier out of the box. Pro adds impact analysis, version control, unit testing, and advanced masking. Enterprise adds audit logging. See Licensing & Tiers for activation, tier comparison, and troubleshooting, or Pricing for plan details.

Quick glossary

• Semantic model: tables/columns/measures/relationships/roles (not report visuals).
• Desktop: local PBIX opened in Power BI Desktop.
• Service (XMLA): Power BI Service dataset connectivity via XMLA endpoint.
• Mode: full vs read-only vs browse-only (org governance / safety).
• Tier: Free / Pro / Enterprise - controls which tools are available.

More references

• Tools index
• Tool capability matrix
• Tool availability matrix
• Licensing & Tiers
• Concepts
• Choose your path

Admin and publishing

If you manage MCP Engine deployments or publish this wiki site:

• Admin home