How to add Runway Dev to Claude Code, Codex and Cursor

Runway Dev now installs into the coding agents a lot of developers already keep open. On November 2, 2026, Runway launched Runway Dev plugins for Claude Code, Codex and Cursor, and the pitch is a single install: the plugin bundles the Runway Dev MCP, which connects your agent to your Runway Dev account, and Runway's skills, which show it how to build with the API.

That means you can describe a feature and let the agent build it without leaving the editor. Runway's own example is a product video for every item in a store. The company lists other jobs in the same shape: an upscale button inside an internal tool, or a localization workflow that runs on new assets.

Here is what the plugin installs, the exact commands for each agent, and the limits worth knowing before you run them.

What the Runway Dev plugin actually installs

The plugin is two parts: the Runway Dev MCP and seven skills. The MCP connects the agent to your Runway Dev account. The skills teach it to build with Runway's official SDKs for Node.js and Python, and which Runway Dev tool fits each job.

Runway's install guide lists the seven skills by name: runway-dev-setup for connecting your account and project, runway-dev for current docs, credentials, SDKs and errors, runway-dev-models for direct model endpoints, then runway-dev-model-routers, runway-dev-characters, runway-dev-recipes and runway-dev-workflows.

The plugin does not generate media. Runway's guide says generations run in your app through the Runway API, which is the part your users actually touch.

How to install it in Claude Code, Codex and Cursor

Each agent takes one install, and the plugin itself is free. Runway's launch post gives the command per agent, and the install guide repeats them.

• Claude Code: inside a session, type /plugin install runway-api@claude-plugins-official. From a terminal, run claude plugin marketplace add anthropics/claude-plugins-official, then claude plugin install runway-api@claude-plugins-official, then claude mcp login plugin:runway-api:runway-dev.
• Codex: run codex plugin marketplace add runwayml/skills, then codex plugin add runway-api@runway-dev, then codex mcp login runway-dev.
• Cursor: type /add-plugin runway-api in Agent chat and select Install, or install from the Cursor Marketplace.

The sign-in command in each case opens a browser to your Runway Dev account. You complete that sign-in yourself, and Runway asks that you not hand the login to the agent.

Two setup notes matter. Runway says to install from one place only, so the agent does not load two copies of the skills. And the Codex IDE extension does not support plugins, so Codex users install in the CLI.

What your agent can read from your Runway Dev account

The MCP reads your account, not a generic catalog. When the agent sets up, it reads the projects available to you, the models each project can use with their prices and limits, and your credit balance. Runway says the integration is sized to what the project actually has access to.

That matters because the model list is not one lab's. Runway Dev carries models from Runway and other frontier labs, and the full set is in the model catalog. Runway's own agent page shows an agent proposing Aleph 2.0, Runway's video editing model, with Seedance 2.5, Kling O3 Standard and Gemini Omni Flash 1.1 as alternatives, and each request sends the inputs its chosen model accepts.

What it builds: Model Routers, Characters, Recipes and Workflows

The agent configures four Runway Dev tools from inside the editor. A Model Router picks the right model for each request, ranked on Runway's quality, price and speed data, and your agent can look up which model the router picked for any call. A Character sets up a real-time avatar, and the agent can list and inspect the ones a project already has.

For common jobs the agent builds on Recipes, Runway's ready-made pipelines. For anything your team has already wired up, it builds on Workflows published as API endpoints in the Runway app.

The pattern in Runway's demo is a plain instruction with a cap. One example on the agent page shows a user asking for fast ad variations that never spend more than $3 on a video, and the agent configuring the router to hold that line.

How it handles a failed task

When a generation fails, the agent looks up the error instead of guessing. Runway's MCP post says the tool returns the specific reason: a moderation rejection, an asset size limit, a malformed request body. The agent also sees why a Model Router picked a given model, so it knows what to check next. Then it fixes the request and runs it again.

That debugging loop is the part most integrations do by hand. Runway frames it as the agent owning the integration from the first development request to live production errors.

What it costs, and the API key rule

The plugin is free. Generations are not. The install costs nothing, and every generation your app runs goes through the API and spends credits at each model's listed price. Runway's pricing page describes one credit balance across the models.

The security detail worth repeating: no API key is stored in the plugin. Runway's post says it plainly, and the MCP guide and the skills repository both say the connection uses browser OAuth at https://dev.runwayml.com/mcp. Your app keeps its own key on your server as RUNWAYML_API_SECRET, and Runway says to keep it out of browser bundles, chat and source control.

What it does not do

The plugin will not create media on its own, so anything you ship still calls the API from your app. New Characters cannot be created through the Claude-specific MCP endpoint, which drops that tool; Runway says to create those in your Runway Dev account.

The version can also differ by marketplace. The skills repository notes that the runway-api@claude-plugins-official listing is managed by Anthropic and may pin a different version from the Runway-maintained marketplace, so if you want Runway's own build you can install from runwayml/skills instead.

Runway Dev reaches beyond these three agents. The docs cover Grok Bot, the repository covers Replit, and any agent that supports MCP and Agent Skills can set up the same two parts with npx skills add runwayml/skills. Runway's agent page adds a quickstart menu that opens Claude Code, Codex or Cursor with a message to the agent already typed in, so it can install the SDK and make the first call.

Runway pitches the platform as enterprise-grade: SOC 2 Type II compliance, IP indemnification, built-in content moderation and 99.9% uptime. For a developer wiring generative media into a product, the change that matters is smaller and more practical. The setup, the model choice and the debugging now happen where the code does.

To start, install the plugin for your agent, sign in to Runway Dev in the browser when it asks, and describe the feature you want built.