# Power BI Authoring MCP Server: AI Agents for Semantic Model Development

> The Power BI Authoring MCP Server is a local MCP server from Microsoft that lets AI agents read and modify Power BI semantic models through natural language, supporting Power BI Desktop, Fabric workspaces, and PBIP project files.

**microsoft/powerbi-modeling-mcp** — The Power BI Modeling MCP Server, brings Power BI semantic modeling capabilities to your AI agents.

- Repository: https://github.com/microsoft/powerbi-modeling-mcp
- Stars: 1,192 · Forks: 211
- Language: Unknown
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/microsoft-powerbi-modeling-mcp

## What Power BI semantic modeling tasks this server automates

Power BI semantic models contain tables, columns, measures, relationships, and security rules that developers typically manage through Power BI Desktop's UI or manual TMDL file edits. The Power BI Authoring MCP Server implements the Model Context Protocol specification to expose those modeling operations to AI agents through a structured interface. An agent can receive a natural-language instruction, translate it into modeling operations, and execute them against a connected model.

The README lists the key capabilities: creating and modifying tables, columns, measures, and relationships; executing bulk operations across hundreds of objects (bulk renaming, bulk refactoring, model translations, security rules); evaluating and implementing modeling best practices; and querying and validating DAX expressions. The server also supports TMDL and Power BI Project (PBIP) files, which allows agentic workflows to operate on model source files as code rather than through a live connection. The project reached version 1.0.0 on 2026-09-25 after a public preview period.

## Three connection modes: Desktop, Fabric, and PBIP

Before issuing any modeling command, the agent must connect to a Power BI semantic model. The README documents three connection prompts depending on where the model lives.

For a model open in Power BI Desktop, use:

```
Connect to '[File Name]' in Power BI Desktop
```

For a model in a Fabric workspace:

```
Connect to semantic model '[Semantic Model Name]' in Fabric Workspace '[Workspace Name]'
```

For a model stored in PBIP files on disk:

```
Open semantic model from PBIP folder '[Path to the definition/ TMDL folder in the PBIP]'
```

Each connection mode covers a different development context. Desktop connections work for local development. Fabric workspace connections require valid workspace credentials. PBIP connections work directly with the TMDL folder on disk, making them compatible with git-based workflows where the model is stored as source files. The README warns that the server can only execute modeling operations and cannot modify report pages, semantic model diagram layouts, or other non-modeling Power BI metadata.

## Installing via VS Code, NPX, or VSIX

The recommended installation is through the Power BI Authoring MCP Visual Studio Code extension, which also requires GitHub Copilot Chat. Once the VS Code extension is installed and Copilot Chat is open, the powerbi-modeling-mcp tool should appear in the available tools list. The README notes that enterprise accounts have the MCP servers in Copilot option disabled by default and an administrator must enable it.

For other MCP clients, the server can be added via NPX without a manual download. Add this JSON configuration to your MCP client:

```json
{
	"powerbi-authoring-local": {
		"type": "stdio",
		"command": "npx",
		"args": [
			"-y",
			"@microsoft/powerbi-modeling-mcp@latest",
			"--start"
		]
	}
}
```

This requires Node.js. For environments without Node.js, a manual VSIX download is also available from the Visual Studio Marketplace. The download URL takes a version number and a target platform (for example win32-x64). After downloading, rename the .visx file to .zip, unzip it, run powerbi-modeling-mcp.exe, and copy the MCP JSON registration from the console output.

## Remote hosted server for Fabric workspace development

The README includes an important note for Fabric workspace users: Microsoft provides a remote (hosted) version of the Power BI Authoring MCP server that requires no local installation. Microsoft manages updates to the remote version. The README links to documentation at learn.microsoft.com for comparing remote and local options and describes the remote server as the preferred choice when authoring semantic models in a Fabric workspace.

The local server described in this repository is the alternative for developers who cannot use the hosted version, or who work with Power BI Desktop or PBIP files. Both serve the same MCP surface. For teams that already have an MCP client set up and want to get started without any server configuration, the remote option avoids the Node.js dependency and the version management overhead of the local server.

## Model quality and safety considerations

The README contains several warnings that reflect real limitations of LLM-driven modeling. First, the AI model choice directly affects result quality. The README recommends deep-reasoning models such as GPT-5 or Claude Sonnet 4.5 for best results. Smaller or instruction-following models may produce inaccurate DAX, incorrect relationship configurations, or incomplete bulk operations.

Second, the README explicitly warns to always create a backup of your model before performing any operations, because the LLM may produce unexpected or inaccurate results leading to unintended changes. Third, LLMs may unintentionally expose sensitive information from the semantic model, including data or metadata, in logs or responses, and the README advises caution when sharing chat sessions. The Data Privacy section of the README covers this further. These are not edge cases; they are documented constraints of the current Public Preview state of the server.

## What the server cannot do

The Power BI Authoring MCP Server is scoped strictly to semantic modeling operations. The README states it cannot modify report pages, visual settings, or semantic model diagram layouts. It also cannot modify other types of Power BI metadata outside the semantic model layer. This means an agent cannot use the server to build or rearrange a report, change visual formatting, or manage Power BI service configurations.

Bulk operations rely on the LLM to interpret natural-language instructions correctly. An instruction to rename hundreds of columns according to a naming convention requires the agent to generate correct names for each object. The README describes this as turning hours of repetitive work into seconds, but the accuracy of bulk results depends on the agent's ability to apply the convention consistently. DAX query validation is available, which helps catch calculation errors, but it does not guarantee that a generated measure is semantically correct for the intended analysis.

## A comparison with the Analysis Services XMLA endpoint approach

Before MCP servers existed for Power BI, developers who wanted to automate semantic model changes typically used the Analysis Services XMLA endpoint directly, which requires SSAS client libraries or Tabular Editor with its scripting API. XMLA-based automation requires knowledge of the Tabular Object Model (TOM) and C# or scripting, making it inaccessible to developers who know DAX and Power Query but not .NET.

The Power BI Authoring MCP Server moves the interface to natural language via an AI agent. The tradeoff is precision: a developer using TOM scripting has deterministic control over every property. An AI agent interpreting a natural-language instruction introduces a layer of translation that can introduce errors, which is why the README recommends backups and caution. For teams that already use Tabular Editor for complex model automation, the MCP approach offers less precision but far lower scripting overhead. For teams without existing automation, the MCP server is an accessible starting point.

## Maintenance status and license

The repository's last push was on 2026-09-25, the same day as the v1.0.0 release. The project is not archived and carries an MIT license. The README notes the project is in Public Preview and implementation may significantly change prior to General Availability. This means the API surface, connection behavior, and supported operations are all subject to change before the project is declared generally available.

The repository root contains CODE_OF_CONDUCT.md, SECURITY.md, SUPPORT.md, and TROUBLESHOOTING.md, following standard Microsoft open-source repository conventions. EULA.txt is also present, which is separate from the MIT license. Teams integrating the server into production workflows should review both the MIT license and the EULA before deployment, as the EULA may impose additional terms specific to the extension components.

## Conclusion

The Power BI Authoring MCP Server is the right choice for Power BI developers who want to drive semantic model changes through an AI agent rather than the Power BI Desktop UI. It is not suitable for modifying report pages, diagram layouts, or any Power BI artifact outside the semantic model layer. Before connecting an agent, create a backup of your model: the README explicitly warns that LLM output may be inaccurate and can cause unintended changes. For Fabric workspace users, Microsoft recommends the remote hosted version over the local server, since it requires no installation and Microsoft manages updates.

## FAQ

### How do I install the Power BI Modeling MCP server VSIX?

Download the VSIX file from the Visual Studio Marketplace using the URL template in the README, which takes a version number and platform (such as win32-x64). Rename the downloaded .visx file to .zip, unzip it, run powerbi-modeling-mcp.exe, and copy the JSON registration from the console to configure your MCP client.

### Does the Power BI Authoring MCP Server require Node.js?

The NPX installation method requires Node.js, as it downloads and runs the @microsoft/powerbi-modeling-mcp npm package. The VS Code extension method and the manual VSIX download do not require Node.js separately.

### Can the Power BI Authoring MCP Server modify report pages?

The README explicitly states the server can only execute modeling operations and cannot modify report pages or other Power BI metadata such as diagram layouts. It is limited to the semantic model layer.

## Sources

- [Issues](https://github.com/microsoft/powerbi-modeling-mcp/issues)
- [License: MIT](https://github.com/microsoft/powerbi-modeling-mcp/blob/main/LICENSE)
- [microsoft/powerbi-modeling-mcp on GitHub](https://github.com/microsoft/powerbi-modeling-mcp)
- [README](https://github.com/microsoft/powerbi-modeling-mcp/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/microsoft-powerbi-modeling-mcp