Runway MCP workflows

Not for: one-shot generate_image / generate_video — use the runway-media skill. Not for: published Workflow Apps.

Domain knowledge (port types, node catalog, model modes, worked examples) lives in the server's MCP resources, which are always current. Do not copy tables into this skill.

Before you start

1. Runway MCP connected, workspace selected.
2. Read the resources: runway://docs/workflows/overview, then authoring, models, catalog, and examples.
3. list_workflows → get_workflow with include_graph: true on a similar template. Copy that graph shape; do not invent { type, config, inputs } or human-readable ids like input_photo.
4. Sketch the pipeline as a mermaid flowchart LR and show it to the user with inputs, node count, and estimated credits before writing JSON.

Author and run

• New workflow → create_workflow. Edit an existing one → save_workflow_version with the API graph JSON copied from get_workflow (graphs must use UUID ids).
• validate_workflow_graph before saving when unsure; the save still succeeds with errorCount > 0, so surface errors to the user.
• Run with run_workflow. It returns a task id; say generation is in progress. The inline viewer polls where available — do not sleep + get_task in a loop unless the authoring guide says to.
• Never tell the user to build the graph themselves in the editor when the tools can do it.