Runway plugs its creative models into Claude, ChatGPT, Cursor, and Replit with MCP
The new connector makes image and video generation callable inside agent workflows and exposes models like Gen-4.5, Kling, and GPT image 2 right from chat.
By Ryan Merket · Published
Why it matters
Agents are becoming the primary UI for work. By making video and image generation callable from those agents, Runway shifts creative AI from destination apps to infrastructure developers can compose.
Runway quietly shipped an MCP connector that lets people generate images and videos from inside Claude, ChatGPT, Cursor, and Replit, positioning creative AI as callable infrastructure in agent workflows, according to a thread on X and the new Runway MCP page.
https://x.com/runwayml/status/2059636517283176479?ref=runtimewire
Runway has spent the past year arguing that language models alone will not solve the hardest problems in AI, and that simulation via world models is the path forward. Its research and products span Gen-4.5 for high-fidelity video, GWM-1 for interactive simulation, and real-time Runway Characters. MCP is the latest step in making those capabilities usable where builders already work.
What shipped
- A connector that brings Runway generation into agent and IDE surfaces. The MCP page lists Claude, ChatGPT, Cursor, and Replit as compatible entry points. Runway MCP
- Access to multiple creative models from chat, including Seedance 2.0, GPT image 2, Kling, Nano Banana Pro, and Runway’s own Gen-4.5, all exposed behind one connector, per the product page. Runway MCP
- Promptable workflows: generate and restyle product shots, turn a product URL into a launch video, and script multi-shot, dialogue-driven ads without leaving the conversation. Runway MCP
To set it up in Claude, Runway provides a three-step flow: go to Customize -> Connectors, add a custom connector named Runway with the URL Runway MCP endpoint, then connect and sign in with a Runway account. After that, you can ask the agent to, for example, make an ad video. The company has a short setup guide. Connecting to Runway MCP
Runway MCP generated product image example
Why this fits Runway’s arc
Runway describes its mission as building AI to simulate the world. GWM-1 is framed as a general world model with variants for explorable environments, conversational avatars, and robotics. GWM-1 Paired with Gen-4.5 for cinematic motion and visual fidelity, the company has been moving from a standalone creative app to a stack of controllable, real-time systems. MCP extends that trajectory by dropping those capabilities into the agent interfaces and developer tools people already use.
The product page positions MCP as bringing "a complete generation studio, inside your agent" and shows examples of chat-native workflows for marketers, product teams, and creators. By making creative models callable like any other tool, Runway is betting that agents will orchestrate production, with world models and video generation acting as backends rather than destinations.
How to think about it if you build with agents
If you already prototype in Claude or pair program in Cursor or Replit, MCP reduces the friction of hopping out to separate creative tools. It also centralizes model choice behind one connector, which can matter as teams mix aesthetics and capabilities across models like Gen-4.5 and Kling in a single flow. The documentation today shows Claude setup specifically; the MCP page also lists ChatGPT, Cursor, and Replit as compatible.
Runway’s broader push includes partnerships to scale its stack, such as its NVIDIA collaboration to accelerate world and video models, and work with studios like Lionsgate to explore AI in production. Those collaborations hint at where an agent-first creative pipeline could land once the models are embedded in everyday tools. Runway x NVIDIA Lionsgate partnership