Runway launches MCP server for its developer API
Runway has launched Runway Dev MCP, a hosted server that lets coding agents choose generative-media models, configure routing and debug failed jobs on its API platform without leaving the agent.
Runway announced Runway Dev MCP, a hosted Model Context Protocol server. It connects a Runway Dev account to coding tools including Claude, ChatGPT, Codex and Cursor. Developers add it as a custom connector at https://dev.runwayml.com/mcp and approve access.
Runway Dev is the company's API platform for generative media, offering models from Runway and other frontier labs. Through the MCP server, an agent can look up which models a project has access to, along with their prices, capabilities and accepted inputs. It can then pull the exact request schema for the model it picks, so its first call is formatted correctly.
Agents can also create and configure a Model Router that optimizes for cost, latency or quality across several models. The agent can suggest which models should be eligible and set a per-generation cost cap. Runway says routed calls are matched against those rules and then ranked using its own quality, price and speed data. Agents can check which model handled a given call. The server also covers Characters: agents can create them, inspect existing ones and attach or detach the knowledge documents behind them.
For debugging, an agent can look up a failed task and get the specific reason, such as a moderation rejection, an asset size limit or a malformed request body. It can then fix the request and run it again. Runway says the server is more token-efficient, but it gave no figures.
The launch comes alongside documentation written for agents to read as context, plus a Quickstart menu that opens Claude Code, Codex or Cursor with a starter prompt already filled in. Runway did not say whether the MCP server costs anything extra.