Claude: Connect to Microsoft Fabric

Key insights

• Fabric Core MCP Server: A Microsoft-hosted MCP endpoint that lets AI assistants talk to Fabric over HTTP.
  Use the official endpoint https://api.fabric.microsoft.com/v1/mcp/core to connect Claude or other MCP-compatible hosts without installing local server software.
• MCP (Model Context Protocol): A standard API surface that exposes Fabric resources, lets hosts discover capabilities, and supports streamable HTTP.
  Claude, Claude Code, VS Code and other MCP hosts use this protocol to authenticate and call Fabric services.
• Microsoft Entra ID and auth flow: You must register an app or complete the browser OAuth flow so Claude can act for your tenant.
  Set the auth scope to https://api.fabric.microsoft.com/.default and finish sign-in to grant access.
• Redirect URI and permissions: Setting the correct redirect URI during app registration is the most common setup error.
  Also configure the required API permissions in Entra ID so Claude can authenticate and call Fabric on your behalf.
• Practical examples: Once connected, prompt Claude to list workspaces, enumerate items inside a workspace (lakehouses, semantic models, reports), or auto-generate documentation for a semantic model and its reports.
  These examples show read, explore and document scenarios; the same connection can enable conversational queries, creation, or modification when permitted.
• Hosted vs local MCP servers: Microsoft’s remote Fabric Core MCP Server removes the need for local installs and provides a straightforward “Add Server” workflow.
  Local or community MCP servers still exist for specialized or offline needs, but the hosted endpoint is the authoritative, easier path for most tenants and works with Claude on the web as well as code editors.

Video summary and context

In a recent YouTube demonstration, Reza Rad (RADACAD) [MVP] walks viewers through a complete setup that connects Claude to Microsoft Fabric using the Fabric Core MCP Server. The video aims to show the end-to-end process, from registering an application in Microsoft Entra ID to authenticating Claude and calling Fabric APIs. Moreover, the presenter highlights practical examples such as listing workspaces and auto-generating documentation, illustrating how the connection changes daily workflows. Overall, the tutorial emphasizes a repeatable configuration that works with both the web and code-based versions of Claude.

Overview of the Fabric Core MCP Server approach

First, the video explains that MCP stands for Model Context Protocol, a standard that lets AI hosts talk to external systems through a structured server interface. Then, it clarifies that Microsoft provides a hosted Fabric Core MCP Server endpoint so MCP-compatible tools can connect over HTTP rather than requiring a local server. Consequently, this hosted option lowers setup friction, especially for users who prefer browser-based authentication and cloud-hosted endpoints. Finally, the presenter notes that the same endpoint supports multiple MCP hosts, including developer tools and web-based assistants.

Step-by-step connection process

Next, Reza lays out the essential steps for making the connection work, starting with app registration in Microsoft Entra ID. He explains how to create an application, set the correct redirect URI, and why an incorrect redirect URI is the single most common configuration error. In addition, he walks through granting the right API permissions and choosing scopes that let the AI authenticate and call Fabric on behalf of users. He stresses careful permission selection to avoid over-privileging the AI application while keeping necessary functionality.

After app registration, the video shows how to add the Fabric Core MCP Server inside Claude settings using the app registration details. Then, viewers see the browser-based OAuth flow that completes the authentication and saves the configuration for future use. Along the way, Reza demonstrates the difference between connecting via a hosted endpoint and using a local community MCP server, pointing out that the hosted endpoint simplifies the process. Finally, he reminds administrators to verify workspace access in Fabric before attempting queries from Claude.

Demonstrations: what Claude can do once connected

In the demo section, Reza asks Claude to list every workspace in a Fabric tenant, showing that the assistant can explore tenant structure without manual portal navigation. Then, he drills down further by prompting Claude to enumerate items inside a specific workspace, including lakehouses, semantic models, and reports. Moreover, he requests automatic documentation generation for a semantic model and a report, demonstrating how the assistant can produce readable documentation from Fabric metadata. These examples underline that documentation is merely a clear demonstration of broader possibilities.

Tradeoffs and security considerations

However, the hosted versus local approach comes with tradeoffs that administrators must weigh carefully. The hosted Fabric Core MCP Server reduces complexity and speeds onboarding, but it centralizes access and relies on Microsoft’s service availability and policies, which may not suit all compliance models. Conversely, local community servers provide more control and customization, yet they require maintenance and secure hosting, increasing operational overhead. Therefore, organizations should balance ease of use against governance, control, and resilience needs.

Security also demands attention to permission scope and governance controls because the same connection that reads metadata can potentially change resources if granted broader rights. Consequently, the video recommends following least-privilege principles and testing with read-only permissions first before expanding capabilities. In addition, conditional access, multi-factor authentication, and tenant policies can affect the OAuth flow, so administrators should validate those settings early. Finally, logging and audit trails remain crucial to track what the assistant accesses or modifies.

Challenges, best practices, and operational tips

Reza highlights common pitfalls and offers practical troubleshooting tips that save time during setup. For example, an incorrect redirect URI stops authentication at the start, and missing API permissions prevent calls even after a successful sign-in, so verifying both items is essential. Moreover, tenant-wide policies such as blocked consent or stricter conditional access may require coordination with identity administrators to permit the connection. As a best practice, he suggests beginning with small, read-only tasks and gradually introducing create or modify operations after confirming governance and change management processes.

Finally, the video underscores the potential for expanding conversational capabilities beyond documentation to querying, creating, and modifying Fabric objects, but it also cautions about operational risk. Teams should build change approvals, testing sandboxes, and monitoring before enabling write operations at scale. In summary, Reza’s demonstration provides a practical road map that balances innovation with responsible implementation, giving organizations a clear starting point to evaluate the benefits and risks of connecting Claude to Microsoft Fabric.

Keywords

Connect Claude to Microsoft Fabric, Fabric Core MCP server setup, Anthropic Claude Microsoft Fabric integration, Microsoft Fabric Claude tutorial, Claude API Fabric connector, Fabric MCP server configuration guide, Integrate Claude with Fabric analytics, End-to-end Claude Fabric setup