runway mcp starts a Model Context Protocol (MCP) server, allowing AI tools like Claude Code, Cursor, and others to interact with your Runway apps directly.
Configure your AI tool to connect to the runway mcp server, using stdio
transport. For example, to add it to Claude Code:
claude mcp add --scope user --transport stdio runway -- runway mcp
By default, all tool categories are enabled, and the MCP server can do anything
that the runway CLI tool can do. Use the following flags to restrict what the
MCP server exposes:
| flag | default | description |
|---|---|---|
--[no-]write |
on | Expose non-readonly tools |
--[no-]destructive |
on | Expose destructive tools (that may delete or overwrite data) |
--[no-]open-world |
on | Expose tools with open-world (arbitrary code execution) access, like runway app exec |
--[no-]deployment |
on | Expose deployment tools |
--deployment flag only makes sense when the AI tool has access to your local project directory, such as when using Claude Code or Cursor inside your repo.To give an AI tool a read-only view of your apps, use:
runway mcp --no-write --no-destructive --no-open-world --no-deployment