Setting up Fabric MCP Server

This guide walks through setting up the Fabric MCP Server (local) on a development machine. The server runs as a subprocess and connects an MCP-capable AI agent to Fabric API documentation, OneLake file operations, and item creation tools.

Prerequisites

Check the following before installing:
MCP-compatible AI client (VS Code with GitHub Copilot Chat is recommended)
For npx: Node.js 20 LTS or later (node --version)
For source build: .NET 9 SDK (dotnet --version)
Microsoft Fabric credentials (for OneLake operations and item creation)

Option 1: VS Code extension (recommended)

Fastest setup with no manual configuration:
Install the GitHub Copilot Chat extension in VS Code
Install the Fabric MCP Server extension from the VS Code Marketplace
The extension registers the server automatically — no further configuration needed
Switch to Agent mode in GitHub Copilot and refresh the tools list

Option 2: npm/npx (any MCP client)

Add the following to the client's mcp.json file:

{
    "mcpServers": {
        "fabric-mcp-server": {
            "command": "npx",
            "args": [
                "-y",
                "@microsoft/fabric-mcp@latest",
                "server",
                "start",
                "--mode",
                "all"
            ]
        }
    }
}

When configuring VS Code manually, use servers instead of mcpServers as the root key.

Option 3: Build from source (.NET)

git clone https://github.com/microsoft/mcp.git
cd mcp
dotnet build servers/Fabric.Mcp.Server/src/Fabric.Mcp.Server.csproj --configuration Release

Then add to mcp.json:

{
    "mcpServers": {
        "Fabric MCP Server": {
            "command": "/path/to/repo/servers/Fabric.Mcp.Server/src/bin/Release/fabmcp",
            "args": ["server", "start"],
            "type": "stdio"
        }
    }
}

On Windows, use backslashes and add the .exe extension.

Verify your installation

After installing, switch to Agent mode in GitHub Copilot, refresh the tools list, and use this test prompt:

What Fabric workload types are available?

If the agent returns a list of Fabric workload types, the server is working correctly.

Example prompts

After a successful installation, try these example prompts:
"Show me the OpenAPI operations for 'notebook' and give a sample creation body"
"Show me the JSON schema for a Data Pipeline item definition"
"List my Fabric workspaces"
"Download the config.json file from my Sales lakehouse"
"Create a new Lakehouse called CustomerData in my dev workspace"

Security guidance

Protect credentials — Never store access tokens or local Fabric credentials as plaintext in configuration files. Use short-lived tokens with automatic renewal and separate development identities.

Scope file system access — Configure the server only for explicitly required directories; do not use unrestricted path wildcards.

Account for prompt injection — API documentation and OneLake file contents can contain crafted control directives. Review file contents from unknown sources before agent processing.

Use a development workspace — Isolate item creation operations in a dedicated development workspace to avoid unintended changes to production data.

Understand monorepo stars — The GitHub star count belongs to the entire microsoft/mcp monorepo, not the Fabric server component in isolation.

Troubleshooting

Server does not appear in the tools list — Switch to Agent mode in GitHub Copilot and press the refresh button on the tools list. If the server still does not appear, restart VS Code and verify the extension is enabled in the Extensions panel.

Node.js version too old — Run node --version; the result must be v20.x.x or higher. Download and install Node.js 20 LTS from the official website.

OneLake authentication error — Verify that the credentials used have sufficient Fabric permissions. For OneLake read operations, the Viewer role in the target workspace is the minimum; for write operations, Contributor or higher is required.

Manual VS Code configuration fails — Check that your mcp.json uses servers instead of mcpServers as the root key. Microsoft documents this difference explicitly for VS Code configurations.

Tips for daily development

The Fabric MCP Server (local) pairs well with the Fabric Core MCP Server (remote): use the local server for API documentation and OneLake file operations, and the remote server for workspace management and permissions. Both servers can be active simultaneously in the client configuration.