adf-mcp-server

List all pipelines in the configured ADF factory.

Get the full JSON definition of a specific pipeline.

Query pipeline runs in a time window. Default window is the last 24 hours, ordered most-recent first. Returns a continuationToken when more results are available; pass it back as continuation_token to fetch the next page.

Get full details for a single pipeline run by ID. Use this when you already know the runId and want everything ARM returns about it.

Query activity runs for a specific pipeline run. Use this to drill into which activity failed and read the error message. Activity input/output payloads are truncated by default to protect context; pass full=true to receive them untruncated.

List all triggers in the factory and their runtime state.

List all linked services in the factory (databases, storage accounts, etc).

List all datasets in the factory and the linked service each one belongs to.

List all integration runtimes (IRs) in the factory and their type/state. Useful for spotting offline self-hosted IRs that cause pipeline failures.

List all Data Factory v2 instances visible in the current subscription (derived from ADF_FACTORY_RESOURCE_ID). Useful for discovering the ARM IDs of other factories.

Kick off a new run of a pipeline. Returns the new runId. WRITE OPERATION: this consumes ADF resources and may incur cost.

Cancel an in-progress pipeline run. By default cancels child runs as well.

Re-execute a previous pipeline run. By default resumes from the failed activity (the common 'fix and retry' workflow); set from_failed_activity=false to replay from the beginning.

Start a trigger so it begins firing per its schedule. Long-running; verify the resulting runtimeState with list_triggers.

Stop a trigger so it stops firing. Long-running; verify the resulting runtimeState with list_triggers.

Create a new pipeline or overwrite an existing one. Two-step: first call (dry_run=true, default) returns a diff and a confirm_token; second call (dry_run=false with that token) applies the change.

Delete a pipeline. Two-step: first call (dry_run=true, default) returns the resource that will be removed and a confirm_token; second call (dry_run=false with that token) deletes it.

Create or overwrite a trigger. Same dry_run/confirm_token pattern as create_or_update_pipeline. NOTE: a newly-created trigger is in Stopped state — call start_trigger to activate it.

Delete a trigger. Must be stopped first (use stop_trigger). Same dry_run/confirm_token pattern as delete_pipeline.

Create or overwrite a linked service (database / storage connection). Same dry_run/confirm_token pattern. Secrets in connection strings should be referenced from Key Vault, not inlined.

Delete a linked service. Any datasets that reference it will start failing — check with list_datasets first. Same dry_run/confirm_token pattern.

Create or overwrite a dataset. Same dry_run/confirm_token pattern.

Delete a dataset. Any pipelines that reference it will start failing — check with list_pipelines first. Same dry_run/confirm_token pattern.