Runway Dev Plugins for Claude Code, Codex and Cursor: Setup, Pricing and Limits

Runway launched Dev plugins for Claude Code, Codex and Cursor on November 2. They connect a coding agent to a Runway Dev account and add instructions for building media features through its API. The useful change is access to the account's actual capabilities while the agent writes code.

For a founder adding product videos to an app, that can shorten the gap between a plausible integration and one that matches the project's available models, budget and inputs. It also introduces permissions and billing questions that are easy to miss in a one-install announcement.

What the Runway Dev plugins add

The launch announcement combines two components: Runway Dev MCP, which connects the agent to account tools, and skills that guide implementation. The agent can inspect projects, model availability, prices, limits and credit balance. It can also investigate failed tasks and model-routing decisions.

This extends an existing developer interface. Runway introduced its Dev MCP in September. The November release packages that access with implementation guidance for three coding environments, rather than introducing the underlying connection for the first time.

The practical difference is that an agent can work with project-specific constraints instead of relying entirely on documentation or remembered API behavior.

How setup works

Runway's announced installation routes are:

• Claude Code: /plugin install runway-api@claude-plugins-official
• Codex: codex plugin marketplace add runwayml/skills, followed by codex plugin add runway-api@runway-dev
• Cursor: /add-plugin runway-api in Agent chat.

After installation, connect a Runway Dev account through the sign-in flow. Installation alone does not create the account or configure the application's API key.

The public plugin repository documents a separate Runway-maintained Claude marketplace. It also notes that Anthropic's official listing may pin a different version. If the installed behavior differs from a tutorial, check the marketplace and plugin version before changing application code.

Plugin pricing and API billing are separate

The plugin is free to install; generation remains a paid API operation. Runway's coding-agent documentation says the MCP does not generate media, create API keys or create projects. The application makes generation requests using its own key, kept in its server environment.

That distinction matters when testing. Connecting an editor is not a pool of free video credits, and the plugin holding no API key does not remove the application's need to manage one safely.

Model-specific prices determine generation costs. For a product-video feature, estimate the full workflow, including retries and variations, before making it available to every customer. A working integration and an affordable product feature are separate milestones.

What the skills do, and where Claude differs

Seven skills cover setup, the core integration, direct model calls, Model Routers, Characters, Recipes and Workflows. Official Node.js and Python SDKs are supported. These instructions help the agent select the appropriate API surface rather than treating every media task as the same generation request.

There is an important host-specific limit. The repository says Claude's MCP endpoint omits the avatar-creation tool. New avatars must be created in the Developer Portal. Claude's skills also direct validation toward local checks, leaving live generations and workflow execution to the application.

Do not assume every advertised platform capability is exposed identically in every host. That is particularly relevant when moving an integration between coding agents or debugging an action that the underlying platform supports but a plugin does not expose.

Who should use it

This is most relevant to teams already building image, audio or video features into software. A product engineer can give the agent more useful context about the account without repeatedly copying configuration details between a portal and an editor.

The remaining work is still product engineering: deciding who can trigger generation, reviewing server-side key handling, setting spend limits and testing failure paths. The launch provides no independent evidence that an agent-written integration is production-ready without that review.

The more useful test for these plugins is whether they reduce configuration mistakes across a real development cycle, from the first implementation through a failed generation and its fix. That is where account-aware tooling can earn its place in an editor.

I'm Paras. I work at the intersection of AI, engineering and distribution.

If you're building an AI product, I help turn technical launches and product thinking into stories developers and founders read and share.

Book a 30-minute call.