microsoft_fabric_mcp

Microsoft Fabric MCP provides a read-only API to explore and analyze Microsoft Fabric resources. It exposes 25 tools across core Fabric management, data and schema operations, job monitoring, and operational intelligence. You can list workspaces, lakehouses, tables, and schemas, inspect item properties and identities, monitor job instances, view schedules, and analyze dependencies and data lineage. The server is designed for read-only access (GET operations only) and uses smart caching to deliver fast responses for repetitive queries. It integrates with MCP-compatible AI tools like Cursor and Claude, enabling natural-language queries to retrieve Fabric metadata and insights.

To use the MCP tools, first connect your MCP client (e.g., Cursor) to the running MCP server. You can fetch lists of workspaces with list_workspaces, inspect a specific workspace with get_workspace, enumerate lakehouses via list_lakehouses, and drill into table schemas with get_table_schema. Operational intelligence tools let you examine compute usage, item lineage, and dependencies. The 25 tools cover resources such as workspaces, lakehouses, tables, jobs, shortcuts, schedules, and environment configurations, enabling comprehensive read-only visibility into your Fabric environment.

Prerequisites:

• Python environment with access to install packages
• Azure CLI installed and configured (az login) if you plan to use Fabric resources that require Azure authentication
• Optional: UV (Universal Viewer) or UVX for MCP tooling

Option A: UVX Installation (Recommended)

1. Install UVX following the instructions in the README (UVX is the recommended approach):

# Install UVX via the recommended script or package manager as per your OS

2. Install the MCP package via UVX by referencing the Microsoft Fabric MCP package name:

uvx microsoft-fabric-mcp

3. Configure Cursor (or your MCP client) to point to the MCP server:

"mcp_fabric": {
  "command": "uvx",
  "args": ["microsoft-fabric-mcp"]
}

Option B: Local Development (manual/dev mode)

1. Clone the repository and install locally in editable mode:

git clone https://github.com/Augustab/microsoft_fabric_mcp
cd microsoft_fabric_mcp
uv pip install -e .

2. Run the MCP server directly (example path; adjust to your environment):

uv run fabric_mcp.py

3. Add to Cursor MCP settings (example):

"mcp_fabric": {
  "command": "uv",
  "args": [
    "--directory",
    "/path/to/microsoft_fabric_mcp",
    "run",
    "fabric_mcp.py"
  ]
}

Note: Both methods run the MCP server locally. The UVX approach generally simplifies installation and maintenance, while the Local Development approach is useful for development and testing before publishing to a package index.

Tips and best practices:

• The MCP exposes 25 tools across core Fabric management, data/schema management, job monitoring, and operational intelligence. Use these to list resources, inspect details, and analyze dependencies and data lineage.
• The server uses caching to speed up responses; if you make structural changes or rename resources, you can refresh cached data with the provided cache-clearing commands in your environment (as documented in the server UI or tooling docs).
• Ensure Azure CLI authentication is working if your queries access Fabric resources that require Azure authentication. Run az login and verify your subscription context with az account show.
• When configuring in Cursor, you’ll typically add an MCP entry named something like mcp_fabric with the appropriate command (uvx or uv) and the package name or entry point for your installation.
• If you encounter permissions or path issues on Windows, use the Windows guidance in the README to create a batch file or adjust PATH settings to ensure uv or uvx is discoverable.