Power your agent
Install from the publisher. Not a PlugMCP endpoint.
Host a copy with our keys and logs.
No packaged release in the registry yet. Follow the repository README.
Remote URL or a command on your machine? How the two installs differ
Run on PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
LaunchA Model Context Protocol (MCP) server for Microsoft Fabric. Exposes Fabric operations (workspaces, notebooks, SQL, Livy, pipelines, jobs) as MCP tools that AI agents can invoke.
⚠️ Warning: This package is intended for development environments only and should not be used in production. It includes tools that can perform destructive operations (e.g.,
delete_item,delete_lakehouse_file,delete_activity_from_pipeline) and execute arbitrary code via Livy Spark sessions. Always review AI-generated tool calls before execution.
The fastest way to use this MCP server is with uvx:
uvx ms-fabric-mcp-server
# Using uv (recommended)
uv pip install ms-fabric-mcp-server
# Using pip
pip install ms-fabric-mcp-server
# With SQL support (requires pyodbc)
pip install ms-fabric-mcp-server[sql]
# With OpenTelemetry tracing
pip install ms-fabric-mcp-server[sql,telemetry]
Uses DefaultAzureCredential from azure-identity - no explicit credential configuration needed. This automatically tries multiple authentication methods:
AZURE_CLIENT_ID, AZURE_TENANT_ID, AZURE_CLIENT_SECRET)az login)No Fabric-specific auth environment variables are needed - it just works if you're authenticated via any of the above methods.
Add to your VS Code MCP settings (.vscode/mcp.json or User settings):
{
"servers": {
"MS Fabric MCP Server": {
"type": "stdio",
"command": "uvx",
"args": ["ms-fabric-mcp-server"]
}
}
}
Add to your claude_desktop_config.json:
{
"mcpServers": {
"fabric": {
"command": "uvx",
"args": ["ms-fabric-mcp-server"]
}
}
}
Add to your Codex config.toml:
[mcp_servers.ms_fabric_mcp]
command = "uvx"
args = ["ms-fabric-mcp-server"]
# Using uvx (no installation needed)
uvx ms-fabric-mcp-server
# Direct execution (if installed)
ms-fabric-mcp-server
# Via Python module
python -m ms_fabric_mcp_server
# With MCP Inspector (development)
npx @modelcontextprotocol/inspector uvx ms-fabric-mcp-server
MCP stdio servers must keep protocol traffic on stdout, so redirect stderr to capture logs.
Giving the agent read access to the log file is a powerful way to debug failures.
You can also set AZURE_LOG_LEVEL (Azure SDK) and MCP_LOG_LEVEL (server) to control verbosity.
VS Code (Bash):
{
"servers": {
"MS Fabric MCP Server": {
"type": "stdio",
"command": "bash",
"args": [
"-lc",
"LOG_DIR=\"$HOME/mcp_logs\"; LOG_FILE=\"$LOG_DIR/ms-fabric-mcp-$(date +%Y%m%d_%H%M%S).log\"; uvx ms-fabric-mcp-server 2> \"$LOG_FILE\""
],
"env": {
"AZURE_LOG_LEVEL": "info",
"MCP_LOG_LEVEL": "INFO"
}
}
}
}
VS Code (PowerShell):
{
"servers": {
"MS Fabric MCP Server": {
"type": "stdio",
"command": "powershell",
"args": [
"-NoProfile",
"-Command",
"$logDir=\"$env:USERPROFILE\\mcp_logs\"; New-Item -ItemType Directory -Force -Path $logDir | Out-Null; $ts=Get-Date -Format yyyyMMdd_HHmmss; $logFile=\"$logDir\\ms-fabric-mcp-$ts.log\"; uvx ms-fabric-mcp-server 2> $logFile"
],
"env": {
"AZURE_LOG_LEVEL": "info",
"MCP_LOG_LEVEL": "INFO"
}
}
}
}
from fastmcp import FastMCP
from ms_fabric_mcp_server import register_fabric_tools
# Create your own server
mcp = FastMCP("my-custom-server")
# Register all Fabric tools
register_fabric_tools(mcp)
# Add your own customizations...
mcp.run()
Environment variables (all optional with sensible defaults):
Copy .env.example to .env and customize as needed.
The server provides 57 core tools, with 3 additional SQL tools when installed with [sql] extras (60 total).
SQL tools require pyodbc and the Microsoft ODBC Driver for SQL Server (Driver 18 or 17 — the service auto-detects which is installed and prefers Driver 18; set FABRIC_ODBC_DRIVER to override):
# Install with SQL support
pip install ms-fabric-mcp-server[sql]
# On Ubuntu/Debian, install the ODBC driver first:
curl https://packages.microsoft.com/keys/microsoft.asc | sudo apt-key add -
curl https://packages.microsoft.com/config/ubuntu/$(lsb_release -rs)/prod.list | sudo tee /etc/apt/sources.list.d/mssql-release.list
sudo apt-get update
sudo ACCEPT_EULA=Y apt-get install -y msodbcsql18 # or msodbcsql17
If pyodbc is not available, the server starts with 57 tools (SQL tools disabled).
# Clone and install with dev dependencies
git clone https://github.com/your-org/ms-fabric-mcp-server.git
cd ms-fabric-mcp-server
pip install -e ".[dev,sql,telemetry]"
# Run tests
pytest
# Run with coverage
pytest --cov
# Format code
black src tests
isort src tests
# Type checking
mypy src
Integration tests run against live Fabric resources and are opt-in. They require a pre-provisioned Fabric workspace, Lakehouse, Warehouse, and (for Copy Activity tests) at least one external-source connection. See Setup prerequisites below before your first run.
To get started locally, copy the example env file:
cp .env.integration.example .env.integration
Then fill in values matching your provisioned resources. The full set of variables (with inline comments grouping by purpose) lives in .env.integration.example; this list is for orientation:
Required for all integration tests:
FABRIC_INTEGRATION_TESTS=1FABRIC_TEST_WORKSPACE_NAME — display name of the test workspaceFABRIC_TEST_LAKEHOUSE_NAME — Lakehouse item in the workspace (used as the destination for Copy Activities)FABRIC_TEST_LAKEHOUSE_SQL_DATABASE — the Lakehouse's SQL endpoint database name (typically same as the Lakehouse name)FABRIC_TEST_WAREHOUSE_NAME — Warehouse item in the workspace (used by SQL DML tests)Required for Copy Activity tests (Pipeline Flow):
FABRIC_TEST_DEST_CONNECTION_ID — Fabric connection ID for the destination Lakehouse (used by all Copy Activity tests)Per-engine source inputs — set all 5 vars of a block to enable that block's Copy Activity test; the test skips with a logged reason if any value is missing.
PostgreSQL source (e.g., VM-hosted Postgres reachable via a gateway):
FABRIC_TEST_POSTGRES_CONNECTION_IDFABRIC_TEST_POSTGRES_SOURCE_TYPE (default PostgreSqlSource)FABRIC_TEST_POSTGRES_SCHEMAFABRIC_TEST_POSTGRES_TABLEFABRIC_TEST_POSTGRES_DEST_TABLE_NAMESQL Server source (e.g., VM-hosted SQL Server, Azure SQL DB):
FABRIC_TEST_SQLSERVER_CONNECTION_IDFABRIC_TEST_SQLSERVER_SOURCE_TYPE (default SqlServerSource)FABRIC_TEST_SQLSERVER_SCHEMAFABRIC_TEST_SQLSERVER_TABLEFABRIC_TEST_SQLSERVER_DEST_TABLE_NAMEOptional inputs (other flows skip cleanly when absent):
FABRIC_TEST_SEMANTIC_MODEL_TABLE, FABRIC_TEST_SEMANTIC_MODEL_COLUMNS, FABRIC_TEST_SEMANTIC_MODEL_TABLE_2, FABRIC_TEST_SEMANTIC_MODEL_COLUMNS_2, FABRIC_TEST_SEMANTIC_MODEL_SCHEMAFABRIC_TEST_DATAFLOW_NAMEaz login): AZURE_TENANT_ID, AZURE_CLIENT_ID, AZURE_CLIENT_SECRETPOWERBI_BASE_URL, POWERBI_SCOPES, POWERBI_API_CALL_TIMEOUT, POWERBI_REFRESH_POLL_INTERVAL, POWERBI_REFRESH_WAIT_TIMEOUTRun integration tests:
FABRIC_INTEGRATION_TESTS=1 pytest -m integration
Or filter to a specific test:
FABRIC_INTEGRATION_TESTS=1 pytest -m integration -k "copy_activity"
Notes:
pyodbc and a SQL Server ODBC driver (Microsoft msodbcsql18 recommended). The CI workflow installs it via apt-get install msodbcsql18.The tests assume the following Fabric / Azure infrastructure is already in place. This setup is one-time per environment and is not part of the test run itself:
A dedicated Fabric workspace on an active capacity. Do not use a production workspace — the tests create, modify, and delete items.
A Lakehouse and a Warehouse in that workspace, with display names matching FABRIC_TEST_LAKEHOUSE_NAME and FABRIC_TEST_WAREHOUSE_NAME. The Lakehouse's SQL endpoint database name (which usually matches the Lakehouse name) goes in FABRIC_TEST_LAKEHOUSE_SQL_DATABASE.
Source databases reachable from Fabric (for Copy Activity tests). Options:
Fabric connections to those source databases (one per source-engine you want to test). Create via Fabric portal → Settings → Manage connections and gateways → New connection. Use the Fabric REST API to read back the GUID for FABRIC_TEST_*_CONNECTION_ID:
TOKEN=$(az account get-access-token --resource https://api.fabric.microsoft.com --query accessToken -o tsv)
curl -sS -H "Authorization: Bearer $TOKEN" https://api.fabric.microsoft.com/v1/connections \
| python3 -c "import json,sys
for c in json.load(sys.stdin).get('value',[]):
print(f\"{c['displayName']:40s} {c['id']}\")"
A Lakehouse-destination Fabric connection (Cloud > Lakehouse). Used for FABRIC_TEST_DEST_CONNECTION_ID. Pipeline Copy Activities target this connection when writing to the destination Lakehouse.
Auth:
az login as a Fabric workspace member is sufficient (the server uses DefaultAzureCredential).AZURE_TENANT_ID / AZURE_CLIENT_ID / AZURE_CLIENT_SECRET. Both ACL grants (workspace member + per-connection "Can use") are required — connection access does not inherit from workspace membership.GitHub Actions (if running the bundled workflows): create an environment named Integration in your fork's repository settings and add every FABRIC_TEST_* and AZURE_* variable above as an environment secret with the same name. .github/workflows/integration-tests.yml lists the canonical secret-name set.
MIT
MCP server for Microsoft Fabric with tooling support for workspaces, Notebooks, Livy, Pipelines, Lakehouses, Warehouses, Jobs and Semantic Models
Tools
No tool list yet. Remote servers get one after the first health check.
FAQ
Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.
No required environment variable is listed on this record.
Ms Fabric runs locally over stdio.
Last checked 2026-10-28. MCP server for Microsoft Fabric with tooling support for workspaces, Notebooks, Livy, Pipelines, Lakehouses, Warehouses, Jobs and Semantic Models