Integrationen

Microsoft Fabric: Integration für KI-Agenten

Run Fabric notebooks and jobs, inspect pipelines, and query data via GraphQL

Verbinde Microsoft Fabric mit Langdock und nutze es im Chat, in Agenten und in Workflows.

35 Actions

Beschreibungen, Actions und Beispiel-Prompts kommen direkt aus dem Produkt und sind dort auf Englisch verfasst.

Actions für Microsoft Fabric

Jede Action ist ein Schritt, den ein Agent oder Workflow in Microsoft Fabric ausführen kann.

  • Cancel item job

    Requests cancellation of a running Fabric item job instance. Cancellation is asynchronous; follow it with Wait for item job until the status is Cancelled, Completed, Failed, or Deduped.

  • Commit to git

    Commits uncommitted Fabric workspace item changes directly to the connected Git branch. This does not open a pull request and can fail if that branch is protected. For a review flow on Azure DevOps, use Create branch, Commit files, and Create pull request there, then Update from git here after merge (for GitHub-backed workspaces, use the GitHub integration). With All, commit every dirty item; prefer Selective when the user named specific items. Confirm the workspace before calling this. Workspace ID must be the exact workspace display name, URL, or UUID.

  • Create notebook

    Creates a new Fabric notebook with inline markdown and Python cells. This is not an upsert; use Update notebook to change an existing notebook. When results should come back to the chat after a run, end the last cell with notebookutils.notebook.exit(json.dumps(result)) so Wait for item job returns it as exitValue; use print or df.show rather than display() for other output that should be readable afterwards with Get job logs. Use a unique display name. Workspace ID must be the exact workspace display name, URL, or UUID; if the user says DE or another nickname, call List workspaces first. Use Delete item when the notebook should not remain in the workspace.

  • Create pipeline

    Creates a new Fabric data pipeline from a list of activities, such as waits, notebook runs, and Spark job definition runs chained with dependencies. Checks activity names, dependencies, and notebook or Spark job references before saving. This is not an upsert; use Update pipeline activity to change settings of an existing pipeline.

  • Create Spark job definition

    Creates a new Fabric Spark job definition from an inline PySpark main file, with optional command-line arguments, default lakehouse, and environment. This is not an upsert; use Update Spark job definition to change an existing one. Python only; JAR, Scala, and R jobs are set up in Fabric.

  • Delete item

    Deletes a Microsoft Fabric item from a workspace. The item is soft-deleted when the item type supports recovery. SQL analytics endpoints cannot be deleted on their own; they go with their lakehouse or warehouse. Confirm the workspace and item name before calling this.

  • Delete item schedule

    Permanently deletes one recurring run schedule from a Fabric pipeline, notebook, or Spark job definition. To pause a schedule instead, use Save item schedule with status disabled.

  • Execute GraphQL mutation

    Executes one GraphQL mutation operation that can change Fabric data. Use Inspect GraphQL schema first to identify the available mutation.

  • Execute GraphQL query

    Executes one GraphQL query operation. Use Inspect GraphQL schema first when the operation's fields, return type, filters, or selection shape are not already known; do not use this action for introspection.

  • Get environment

    Returns the published setup of a Fabric environment: Spark runtime version, compute pool and driver and executor sizes, dynamic allocation, and the published libraries. Read-only; environment changes and publishing are done in Fabric.

  • Get git connection

    Returns whether a Fabric workspace is connected to Git and, if so, the provider, organization, project, repository, branch, folder, and web URL. NotConnected is a valid result. This does not list uncommitted item changes; use Get git status. Fabric Git only syncs the workspace copy. Pull requests, branches, and file commits on the Azure DevOps repo are the Azure DevOps integration. Workspace ID must be the exact workspace display name, URL, or UUID.

  • Get git status

    Lists uncommitted Fabric workspace changes and incoming Git remote changes for a git-connected and initialized workspace, including conflicts. Notebook rows include gitFolder and gitPaths for Azure DevOps Commit files. An empty change list means the workspace is in sync. This call may wait on a Fabric long-running operation. Pull requests are Azure DevOps or GitHub, not this action. Workspace ID must be the exact workspace display name, URL, or UUID.

  • Get item

    Returns metadata for one Fabric item. For lakehouses, warehouses, and SQL databases, also retrieves type-specific properties such as SQL endpoint connection strings when the typed Get API succeeds.

  • Get item definition

    Returns the decoded public definition of a Fabric item, including notebook cells or pipeline JSON the model can read. Omit format for pipelines. Microsoft requires write permission to read a definition, so this action uses Item.ReadWrite.All. Huge notebooks are truncated. This call may wait on a Fabric long-running operation.

  • Get item job

    Returns the current status of one Fabric item job instance in a single check, including the notebook exit value for completed notebook runs and, for finished notebook runs when Fabric provides it, a snapshotUrl for the user (it needs the user's Fabric sign-in; do not fetch it). Text a notebook printed (print, df.show) is in Get job logs; display() output cannot be retrieved. To follow a job started by Run item job until it finishes, use Wait for item job instead of calling this repeatedly.

  • Get job logs

    Returns the Spark driver log (up to the last 12,000 characters of its first 4 MB) for one notebook or Spark job definition run, including text the run printed (print, df.show) and the Python traceback of a failed run. Use this to read what a notebook or Spark job printed, or when a Spark run failed with a generic failureReason. display() output is not in this log; a notebook's exit value comes from Get item job or Wait for item job. Not for pipeline runs; use Query pipeline activity runs for those.

  • Inspect GraphQL schema

    Inspects a Fabric GraphQL API schema before composing a query or mutation.

  • List capacities

    Lists Microsoft Fabric capacities the signed-in user can access as an administrator or contributor. Match an exact capacity name and reuse its UUID when another action needs it. If multiple results are plausible, ask the user which capacity they mean instead of guessing.

  • List GraphQL APIs

    Lists Microsoft Fabric GraphQL APIs in a workspace selected by UUID, Fabric URL, or exact user-provided name. Reuse the returned workspace and GraphQL API UUIDs in schema and execution actions. If multiple results are plausible, ask the user which API they mean instead of guessing.

  • List item jobs

    Lists recent job instances for a Fabric item such as a pipeline, notebook, or Spark job definition. An empty list is a valid result when the item has no recent jobs. Reuse a returned job instance UUID with Wait for item job (running jobs), Get job logs (what a notebook or Spark run printed, or why it failed), or Query pipeline activity runs (pipeline runs).

  • List item schedules

    Lists the recurring run schedules of one Fabric pipeline, notebook, or Spark job definition, including whether each is enabled, its frequency, times, and time zone. Use this before Save item schedule to find an existing schedule ID to change.

  • List items

    Lists Microsoft Fabric data-engineering items in a workspace, optionally filtered by type. Reuse returned workspace and item UUIDs in later actions. Semantic models, reports, and dashboards are omitted; use Microsoft Power BI for those. If multiple results are plausible, ask the user which item they mean instead of guessing.

  • List lakehouse tables

    Lists tables in a Fabric lakehouse using the Microsoft Fabric List Tables preview API. Preview responses can change. Prefer the lakehouse name. This action does not query table rows.

  • List workspaces

    Lists Microsoft Fabric workspaces the signed-in user can access. Match an exact workspace name supplied by the user and reuse its UUID in later actions. If multiple results are plausible, ask the user which workspace they mean instead of guessing.

  • Query pipeline activity runs

    Lists activity runs inside one Fabric data-pipeline job, including which activity failed and why. Use after a pipeline job finishes. lastUpdatedAfter and lastUpdatedBefore are taken from the job instance when available.

  • Run item job

    Starts an on-demand Fabric item job and returns immediately with a job instance handle. Use Pipeline for data pipelines, RunNotebook for notebooks, and SparkJob for Spark job definitions. Do not assume the job finished; follow it with Wait for item job. A notebook that ends with notebookutils.notebook.exit(<string>) returns that value as exitValue; text a notebook printed (print, df.show) can be read with Get job logs, but display() output cannot be retrieved.

  • Run table maintenance

    Starts Delta table maintenance on one lakehouse table: OPTIMIZE to compact small files (with V-Order and optional Z-Order columns), VACUUM to remove old unreferenced files, or both. Runs asynchronously; follow it with Wait for item job on the lakehouse.

  • Save item schedule

    Creates a recurring run schedule for a Fabric pipeline, notebook, or Spark job definition, or changes an existing one when Schedule ID is set. Supports daily, weekly, and monthly runs at fixed times, or every N minutes. An enabled schedule runs the item automatically.

  • Search catalog

    Searches the OneLake catalog the way the Explore search bar does: by item name, description, or workspace name across workspaces the signed-in user can access. Returns data-engineering items and GraphQL APIs with their workspace UUIDs for later actions. Semantic models, reports, and dashboards are omitted; use Microsoft Power BI for those. If multiple results are plausible, ask the user which item they mean instead of guessing.

  • Update from git

    Updates a git-connected Fabric workspace with commits from the connected branch. This can overwrite workspace items. If Get git status shows conflicts, this action refuses unless you pass an explicit conflict policy. Confirm the workspace before calling this. Workspace ID must be the exact workspace display name, URL, or UUID.

  • Update item

    Renames a Fabric data-engineering item (lakehouse, notebook, pipeline, Spark job definition, and similar) or changes its description. Does not change the item's content; use Update notebook, Update Spark job definition, or Update pipeline activity for that.

  • Update notebook

    Replaces or appends cells on an existing Fabric notebook. replace is a full overwrite from this action's cells input only; do not Get item definition and paste that list into replace, because Get is size-bounded and would destroy tail cells. To add code, use mode append (this action GETs the current definition internally). To change existing cells, supply the complete new cell list you intend, not a truncated Get output. If you only have a truncated Get, use append for new cells or refuse. Create notebook is not an upsert. When results should come back to the chat after a run, end the notebook with notebookutils.notebook.exit(json.dumps(result)) so Wait for item job returns it as exitValue; use print or df.show rather than display() for other output that should be readable afterwards with Get job logs. Workspace ID must be the exact workspace display name, URL, or UUID; nicknames such as DE are invalid. Use List workspaces to get the exact name.

  • Update pipeline activity

    Changes settings of one existing activity in a Fabric data pipeline, for example a Wait activity's wait time, a Notebook activity's parameters, or a Copy activity's source query. Merges the given keys into the activity's typeProperties and keeps every other activity and setting unchanged. Use Get item definition first to see the activity names and their current typeProperties. Does not add, remove, rename, or reorder activities. Workspace ID must be the exact workspace display name, URL, or UUID; use List workspaces to get the exact name.

  • Update Spark job definition

    Changes an existing Fabric Spark job definition: replaces its PySpark main file, or sets its command-line arguments, default lakehouse, or environment. Settings and files that are not passed stay unchanged, including reference libraries. Use Get item definition with format SparkJobDefinitionV2 to read the current main file first.

  • Wait for item job

    Waits for one Fabric item job to reach Completed, Failed, Cancelled, or Deduped, checking at the provider's suggested interval for up to about 60 seconds. This is the default check after Run item job or Run table maintenance. Returns the notebook exit value (from notebookutils.notebook.exit) for completed notebook runs and, for finished notebook runs when Fabric provides it, a snapshotUrl where the user can see the rendered cells (it needs the user's Fabric sign-in; do not fetch it). Text the notebook printed (print, df.show) is not returned here; read it with Get job logs. display() output cannot be retrieved. If the wait expires while the job is still running, call this action again instead of assuming success.

Einrichtung und Berechtigungen

Eine Action kann nach deiner Bestätigung fragen, bevor sie ausgeführt wird. In einem Agenten steht jede Action auf „Immer erlauben“ oder „Braucht Bestätigung“.

Alle Integrationen