Skip to main content
Pipelines are saved multi-step workflows built from nodes — remove background, upscale, generate, and more — chained together in a graph. Create pipelines in the Pixelcut web app, then run them from any MCP-compatible agent.

​
Running a pipeline

Use the pipelines tool with action run:
The agent calls pipelines (action: get) to inspect the pipeline, then pipelines (action: run) with inputs mapped to the correct node IDs. Inputs are mapped by node alias. Values can be public URLs, Pixelcut asset IDs, or text — the agent picks the right type based on the node’s input spec.

​
MCP actions

The pipelines tool supports these actions: Use display_pipeline to render an interactive node graph preview in MCP App UIs.

​
Finding pipelines

The agent calls pipelines (action: search) with the relevant query, or list to browse everything available.

​
Monitoring runs

Pipeline runs are async. The pipelines (action: run) call returns a run_id. Track progress with:
The agent calls pipelines (action: get_run) to fetch the run state, including per-node status and outputs.

​
End-to-end example

The agent:
  1. Calls pipelines (search) to find the “Product Cleanup” pipeline.
  2. Calls pipelines (get) to inspect the input nodes.
  3. Calls pipelines (run) with the photo mapped to the right input and collection_id set to the target collection.
  4. Polls pipelines (get_run) until the run completes, then surfaces the resulting assets.
For more on pipeline concepts and the web-based pipeline editor, see the Pipelines user guide.