Run workflows from code and agents

The workflow built in the editor does not change: your script, pipeline or agent fills the same inputs and gets a job back.

An App is a workflow with status ready. Platform Apps are public; your own workflows and Apps are private to their project. Both run the same way.

The same five steps through the API or the Scenario MCP:

The parent job's cuCost reads 0: the cost is the sum of the child jobs in metadata.flow.

1. Get the workflow ID and input names

• Your workflows: in the editor, open ... and choose Copy Workflow ID.
• Public workflows have IDs too. Angle Multiplier is wflow_angle-multiplier.

Each input has a name, such as image1 or text1, which is what you send. Read them from the workflow definition:

curl -s "https://api.cloud.scenario.com/v1/workflows/wflow_angle-multiplier?projectId=$PROJECT_ID" \
  -u "$SCENARIO_API_KEY:$SCENARIO_API_SECRET"

Through the Scenario MCP, ask your agent to call workflow_get; it returns the inputs definition and the editor graph. For Angle Multiplier there is one required input: image1 ("Upload Product", an image).

File inputs take an asset ID, so upload your file first (POST /v1/assets, or upload_asset in the Scenario MCP) and keep the asset_... ID.

2. Quote it with a dry run

A dry run returns a quote for the run without running it. On Enterprise, an IP Detection check requested in the same call is still charged.

curl -s -X PUT "https://api.cloud.scenario.com/v1/workflows/wflow_angle-multiplier/run?projectId=$PROJECT_ID&dryRun=true" \
  -u "$SCENARIO_API_KEY:$SCENARIO_API_SECRET" \
  -H "Content-Type: application/json" \
  -d '{ "image1": "asset_..." }'

A dry run answers with HTTP status 269 and the cost in Compute Units: the status means "estimate", not an error, so do not treat it as a failure. dryRun=api estimates at the API tier.

Through the Scenario MCP, the same quote looked like this on October 12, 2026:

// workflow_run  { "workflow_id": "wflow_angle-multiplier",
//                 "inputs": { "image1": "asset_SstohPMc..." }, "dry_run": true }
{ "creativeUnitsCost": 13.75, "creativeUnitsDiscount": 0, "job": {} }

The quote is in Compute Units (CU), the unit credits are metered in. Budget with it.

3. Run it

Send the same request without dryRun. Inputs go directly in the body, keyed by input name.

curl -s -X PUT "https://api.cloud.scenario.com/v1/workflows/wflow_angle-multiplier/run?projectId=$PROJECT_ID" \
  -u "$SCENARIO_API_KEY:$SCENARIO_API_SECRET" \
  -H "Content-Type: application/json" \
  -d '{ "image1": "asset_..." }'

The response is a workflow job. Its metadata.flow lists each node with its own status and, for model nodes, its own child job. Trimmed from the real Scenario MCP run:

{ "job": { "jobId": "job_BfM1ywYj...", "jobType": "workflow",
  "metadata": { "flow": [
    { "id": "imageGenerator1", "type": "custom-model", "status": "processing",
      "modelId": "model_openai-gpt-image-2-5-sunburst", "jobId": "job_HEmqpWhU..." },
    { "id": "text1", "type": "transform", "status": "success",
      "output": "Generate multiple viewing angles of the product shown in this image..." } ],
    "input": { "image1": "asset_SstohPMc..." } } } }

4. Wait for the result

Poll GET /v1/jobs/{jobId} or use webhooks (see Jobs, polling and webhooks).

In the Scenario MCP, jobs_wait does the polling:

// jobs_wait  { "job_ids": ["job_BfM1ywYj..."] }
{ "status": "completed", "jobs": [ { "jobId": "job_BfM1ywYj...", "jobType": "workflow",
  "status": "success", "assetIds": ["asset_kygugJ6u..."], "cuCost": 0 } ] }

The run took about 23 seconds. The output asset is the six-angle grid shown in Platform workflows and templates.

Through an AI agent

With the Scenario MCP connected, an agent can do all of the above from a plain request:

Run the Scenario App "Angle Multiplier" on the lantern photo I uploaded. Quote the cost
first and wait for my go. Then show me the result and tell me what it cost.

The scenario-workflows skill teaches agents the run procedure; scenario-workflow-authoring covers building graphs. See Scenario MCP tools reference and Scenario's skills.

Answer review steps from code

A run that reaches an Approval node or Slice Items in Choose mode waits until someone answers.

A rejected stand-alone node cancels the run. A rejection inside a Loop iteration cascades to the surrounding loop.

Limits

As documented for the API:

For large batches, queue your inputs and keep at most 10 runs in flight.

If it breaks