On September 22, 2026, Runway added Brand Kits to Runway MCP, its Model Context Protocol server that lets outside agents (in tools like Claude, ChatGPT, Cursor, and Replit) drive Runway’s generation tools directly. With this update, an agent connected through MCP can find a project’s Brand Kit and use its assets as references when generating images or video.
What changed
Brand Kits aren’t new to Runway. They’re the product’s existing way of storing a team’s on-brand assets so generations pull from a consistent source instead of loose uploads. What’s new is that this now works through MCP: an agent doesn’t need a human to manually attach the right logo or palette to a prompt, it can look the kit up and reference it directly. Runway is explicit that this is read-only for now. Creating or editing a Brand Kit still happens inside the Runway product itself, not through the agent.
This arrived alongside other September additions to Runway MCP, including support for several Kling models and a Runway plugin for Grok Bot. Taken together, it’s part of a broader push to make Runway’s generation tools something other agents can call into, rather than a destination you only visit directly.
Why it matters for designers
If your team already keeps a Brand Kit current in Runway, this closes a small but real gap in agent-driven workflows. An agent generating variations or exploring options doesn’t have to be told what “on-brand” means for every single request, because it can just read it from the kit. For teams doing high-volume, AI-assisted asset generation, that’s the difference between “mostly on-brand, if someone remembered to attach the right reference” and on-brand by default.
The read-only limitation is worth noting on its own. An agent can use your brand assets but can’t (yet) update them, which keeps a human in the loop for the one part, what actually counts as “the brand,” that probably shouldn’t be automated anyway.
What to watch out for
We haven’t tested this ourselves yet, so treat the “before you use it” panel in the sidebar as open questions. Worth checking directly: what access an MCP-connected agent actually has to a Brand Kit’s underlying files, and whether that’s scoped any differently than a human user’s access inside the product.
