Fabric CLI

Fabric CLI (fab)

Filesystem-style CLI over the Fabric + Power BI REST APIs. Paths use Workspace.Workspace/Item.ItemType/.... The .ItemType suffix is mandatory on items. Workspaces use .Workspace; hidden roots like .capacities, .connections, .domains, .gateways expose tenant resources.

Version / install: v1.5 went GA March 2026; current is v1.6.x (April 2026+). Install with pip install ms-fabric-cli or uv tool install ms-fabric-cli. Requires Python 3.10, 3.11, 3.12, or 3.13. Pre-installed in Fabric Notebooks (no install step needed when running !fab ... from a notebook cell). Confirm version with fab --version. Canonical per-command reference: microsoft.github.io/fabric-cli. Release notes: github.com/microsoft/fabric-cli/releases.

Interactive (REPL) mode: fab config set mode interactive switches fab to a persistent shell where commands no longer need the fab prefix and cd state survives between calls. Default is command-line mode.

AI assets (v1.5+): The repo's .ai-assets/ folder bundles context/, modes/, prompts/, and skills/ files designed to be loaded by AI coding assistants (Copilot, Claude, Cursor) so the assistant can author fab invocations correctly. These are consumable artifacts, not CLI verbs — there is no fab agent or fab ai subcommand.

Authentication

fab reuses the current Azure CLI session (az login). It auto-acquires tokens for Fabric and Power BI audiences. Storage / OneLake operations via fab api -A storage require the same Az CLI session. fab auth tokens do not work for OneLake storage calls from external tools — acquire via az account get-access-token --resource https://storage.azure.com instead (OneLake's audience is Azure Storage, not the Fabric API).

Path Syntax

Workspace.Workspace/Item.ItemType[/Files|Tables/...]
.capacities/MyCap.Capacity
.connections/conn.Connection
.domains/Analytics.Domain
.gateways/gw.Gateway

Names with spaces or apostrophes work inside double quotes with no escaping: "Claude Code's Workspace.Workspace".

Core Command Reference

Navigation & Discovery

Item CRUD

Access Control

Role values by resource:

Assignment

Labels

Jobs

fab job run auto-selects the right job type per item (notebook / pipeline / semantic model refresh / Spark job) — no manual jobType equivalent to the REST API required.

Tables (Lakehouse)

REST API Passthrough (fab api)

Direct REST access when native commands don't cover the operation.

fab api "<endpoint>"                               # Fabric (default)
fab api -A powerbi "<endpoint>"                    # Power BI
fab api -A storage "<onelake path>"                # OneLake DFS
fab api -A azure "<endpoint>"                      # Azure RM
fab api -X post "<endpoint>" -i '<json>|<file>'    # POST with body
fab api "<endpoint>" -q "value[0].id"              # JMESPath filter

Idiom — extract an ID and chain:

WS_ID=$(fab get "Prod.Workspace" -q "id" | tr -d '"')
MODEL_ID=$(fab get "Prod.Workspace/Model.SemanticModel" -q "id" | tr -d '"')
fab api -A powerbi "groups/$WS_ID/datasets/$MODEL_ID/refreshes" -X post -i '{"type":"Full"}'

Workspace Deployment (fab deploy, v1.5+)

One-command CI/CD that wraps the fabric-cicd Python library. Deploys items from local source folders to a target workspace, with environment-aware parameterization. Use this — not fab api — for code-first workspace promotion from a Git checkout.

fab deploy --config <config_file> [-tenv <env>] [-P '<json-array>'] [-f] [--output_format <fmt>]

Examples:

fab deploy --config config.yml --target_env dev
fab deploy --config config.yml --target_env prod -f
fab deploy --config config.yml --target_env test \
  -P '[{"param1":"value1"}]' -f

The config file resolves source folders, target workspace IDs/names, and per-env GUID replacements through fabric-cicd's parameter model — same $workspace.$id and $items.<Type>.<name>.id tokens documented in the Azure DevOps tutorial. Authenticates via the active fab auth session (interactive or SPN).

When to use fab deploy vs Power BI deployment pipelines: fab deploy is source-of-truth-is-Git (workspace = artifact of the deploy). Power BI deployment pipelines is source-of-truth-is-workspace (dev workspace promotes to test/prod workspaces via the service-side pipeline). They are distinct surfaces — don't mix.

Built-in Azure DevOps task (FabricCLI@1, preview)

There are two ways to run fab in an Azure DevOps pipeline:

1. Install-then-script (the pattern in the fabric-cicd ADO tutorial) — a pip install ms-fabric-cli step followed by a script step that calls fab. Works everywhere, but you own the install/version pinning in every pipeline.
2. Built-in task (preview, March 2026) — the Azure DevOps Pipelines extension for Fabric (marketplace ID ms-fabric.fabric-devops-pipelines) ships fab as a native FabricCLI@1 task. The CLI is auto-provisioned in the pipeline runtime — no pip install step. Auth is via an Azure DevOps service connection to your Fabric tenant (rather than passing SPN env vars into a script). Get started at aka.ms/FabCLI.

- task: FabricCLI@1
  inputs:
    scriptType: 'pscore'          # ps | pscore | batch | bash
    scriptLocation: 'inlineScript'
    inlineScript: 'fab ls'
    FabricCLIVersion: 'v1.2.0'    # pin the CLI version

Prefer the built-in task for new pipelines (less boilerplate, service-connection auth); keep install-then-script when you need a CLI version the task doesn't yet offer, a self-hosted agent, or non-ADO CI. The task is preview — pin FabricCLIVersion and validate before relying on it for prod promotion.

Deployment Pipelines (Power BI service-side)

For the service-side stage-to-stage deployment pipelines (different feature from fab deploy), there are no native verbs — use fab api -A powerbi:

fab api -A powerbi pipelines                                     # user pipelines
fab api -A powerbi admin/pipelines                               # all tenant
fab api -A powerbi "pipelines/$PIPELINE_ID/stages"
fab api -A powerbi "pipelines/$PIPELINE_ID/operations"

fab api -X post "deploymentPipelines/$PIPELINE_ID/stages/$STAGE_ID/assignWorkspace" -i '{"workspaceId":"<ws>"}'
fab api -X post "deploymentPipelines/$PIPELINE_ID/deploy" -i '{
  "sourceStageOrder": 0,
  "targetStageOrder": 1,
  "options": {"allowCreateArtifact": true, "allowOverwriteArtifact": true}
}'

Export / Import Formats

Export output layout:

output/
  Model.SemanticModel/
    .platform
    definition/
      database.tmdl
      model.tmdl
      tables/
  Report.Report/
    .platform
    definition.pbir
    definition/

Critical: fab export does not create intermediate directories. Always mkdir -p <out> first, or it fails with InvalidPath. Never include .platform when making definition API calls directly — it is Git integration metadata. See the fabric-tmdl-api skill.

Non-exportable types: .Dashboard, .SQLEndpoint. (.Lakehouse is now exportable as of v1.6.) Check with fab desc .<ItemType>.

VariableLibrary (v1.6+): Variable Libraries now have full CLI coverage via the standard verbs — mkdir, get, set, rm, ls, export, import, cp, mv — backed by the Variable Library REST APIs. Before v1.6 only the portal could manage them.

DuckDB on OneLake

Fastest way to explore lakehouse / warehouse / SQL Database Delta tables without building a semantic model. Requires duckdb installed and az login active.

INSTALL delta; INSTALL azure;
LOAD delta; LOAD azure;
CREATE SECRET (TYPE azure, PROVIDER credential_chain, CHAIN 'cli');

SELECT * FROM delta_scan(
  'abfss://my-workspace@onelake.dfs.fabric.microsoft.com/MyLH.Lakehouse/Tables/bronze/orders'
) LIMIT 10;

CHAIN 'cli' forces use of Az CLI creds; otherwise DuckDB tries managed identity first and fails on local machines.

Friendly names vs GUIDs:

Always use friendly names for schema-namespaced tables. The .Lakehouse / .Warehouse / .SQLDatabase suffix is required.

Raw files under Files/ via read_csv(), read_json(), read_parquet() with glob patterns (*, **). Cross-item joins in one query — different abfss:// paths per item. Read-only; no write-back to lakehouse tables.

Executing DAX via fab api

cat > /tmp/dax.json << 'EOF'
{
  "queries": [{
    "query": "EVALUATE SUMMARIZECOLUMNS ( 'Date'[Year], \"@Total\", [Total Sales] )"
  }],
  "serializerSettings": { "includeNulls": false }
}
EOF
fab api -A powerbi "groups/$WS_ID/datasets/$MODEL_ID/executeQueries" -X post -i /tmp/dax.json

Temp file avoids the double-escaping trap (bash single-quote + JSON double-quote). DAX rules unchanged: EVALUATE required, single-quote tables, fully-qualify columns.

Common Workflows

# Find all semantic models in a workspace
fab ls "Prod.Workspace" -q "[?type=='SemanticModel'].name"

# Bulk export all reports
mkdir -p /tmp/reports
fab ls "Prod.Workspace" | grep ".Report" | while read item; do
  fab export "Prod.Workspace/$item" -o /tmp/reports -f
done

# Rebind a report to a new model after deployment (v1.5 first-class path)
fab set "Prod.Workspace/Report.Report" -q semanticModelId -i "<new-model-id>"

# Trigger semantic model refresh (v1.5 first-class via fab job, no fab api required)
fab job run "Prod.Workspace/Sales.SemanticModel"                       # synchronous
fab job start "Prod.Workspace/Sales.SemanticModel"                     # fire and forget
fab job run "Prod.Workspace/Sales.SemanticModel" -P refreshType:string=Full

# Equivalent REST passthrough (still valid if you need finer control over the body)
fab api -A powerbi "groups/$WS_ID/datasets/$MODEL_ID/refreshes" -X post -i '{"type":"Full"}'
fab api -A powerbi "groups/$WS_ID/datasets/$MODEL_ID/refreshes?\$top=1"

Gotchas

Reference

• Microsoft Learn: Fabric command line interface (install + login)
• GitHub Pages: Fabric CLI documentation (canonical per-command reference)
• GitHub Pages: fab deploy command reference
• Microsoft Learn: Item management overview (REST APIs fab wraps)
• Microsoft Learn What's New: Fabric CLI v1.5 GA — March 2026 — one-command workspace deploys with fabric-cicd integration, first-class Power BI (rebind/refresh), pre-installed in Fabric Notebooks, Python 3.13, 30+ item types.
• Comprehensive MS Learn link bundle (concept / examples / underlying REST / authentication / lakehouse ops / CI/CD): references/REFERENCE.md

See also

• fabric-cicd skill — the Python library fab deploy wraps: direct FabricWorkspace / publish_all_items usage, parameter.yml model, feature flags
• fabric-rest-api skill — direct REST patterns behind fab api
• fabric-tmdl-api skill — definition payloads for semantic models
• fabric-auth skill — token audiences when acquiring tokens outside az login