fabric-mcp-server

A Model Context Protocol server for Microsoft Fabric. It gives an MCP client (Claude Code, Claude Desktop, etc.) read access to Fabric workspaces, items, pipeline run history, job schedules, deployment pipelines, OneLake storage, and read-only DAX queries against semantic models — plus optional write operations (pipeline/notebook runs, refreshes, Git sync, schedules, stage deployments, item create/delete, and workspace role grants) gated behind a write flag.

It talks to the public Fabric REST API (api.fabric.microsoft.com) and the Power BI REST API (api.powerbi.com), authenticating with @azure/identity so you can run it interactively, headless (device code), via the Azure CLI, or as a service principal.

Features

• Resolve workspaces, items, pipelines, deployment pipelines, and semantic models by display name or GUID — no need to hunt for IDs.

• Continuation-token paging, HTTP 429 retry (honors Retry-After), and long-running-operation polling (handles both 202+Location and 200-with-body) built in.

• Read-only by default; all mutating operations are gated behind an explicit write mode and audit-logged to stderr.

Related MCP server: MCP Server for Power BI

Tools

The write tools are only registered when FABRIC_MCP_MODE=write.

Requirements

• Node.js >= 20

• An Entra (Azure AD) identity with access to the target Fabric workspace(s).

• For execute_dax: Build permission on the target semantic model, and the tenant's "Dataset Execute Queries REST API" admin setting enabled.

Install

git clone <this-repo-url> fabric-mcp-server
cd fabric-mcp-server
npm install

Development

Run the automated regression tests:

npm test

Configuration

All configuration is via environment variables (see .env.example). Set them in your MCP client's env block, or copy .env.example to .env for local debugging.

Auth modes

• interactive — opens a browser; best for desktop/AVD.

• device-code — prints a code + URL to stderr; for SSH / WSL / headless.

• cli — reuses your existing az login session.

• azure-powershell — reuses your existing Connect-AzAccount session; for hosts with Az PowerShell but no az CLI.

• service-principal — non-interactive with client ID + secret.

• managed-identity — for hosting on Azure.

• default — tries env → managed-identity → CLI → browser in turn.

Use with Claude Code / Claude Desktop

Add to your MCP client config (e.g. ~/.claude.json for Claude Code, or claude_desktop_config.json):

{
  "mcpServers": {
    "fabric": {
      "command": "node",
      "args": ["/absolute/path/to/fabric-mcp-server/index.js"],
      "env": {
        "FABRIC_AUTH_MODE": "interactive",
        "AZURE_TENANT_ID": "<your-entra-tenant-id>"
      }
    }
  }
}

To enable pipeline execution, add "FABRIC_MCP_MODE": "write" to the env block.

Examples

Check last night's run of a pipeline:

"Using fabric, how did my_pipeline run in My-Workspace last night?" → list_pipeline_runs → get_pipeline_run on the failed instance.

Validate a measure against a live model:

"Run this DAX against the Production Analytics model in My-Workspace: EVALUATE ROW("Sales", [Total Sales])" → execute_dax returns the row.

Security notes

• No secrets are stored in the repo. Credentials come from environment variables at runtime; .env is git-ignored.

• The write tools (run_pipeline, cancel_pipeline_run, refresh_dataset, update_from_git, update_item_definition, create_schedule, update_schedule, delete_schedule, deploy_stage, run_notebook, create_item, delete_item, add_workspace_role) are only exposed under FABRIC_MCP_MODE=write, and each logs an [AUDIT] line to stderr.

• list_onelake / read_onelake_file call the OneLake DFS API (onelake.dfs.fabric.microsoft.com), which uses the Azure Storage token audience (https://storage.azure.com/.default) — the same @azure/identity credential acquires it, no extra configuration.

• update_item_definition overwrites the live item wholesale; it snapshots the current definition to a local JSON first (under the OS temp dir) and returns the path so you can roll back with restore_snapshot.

• execute_dax uses the Power BI executeQueries API, which only runs read-only DAX (data-modifying queries are rejected by the service).

License

MIT