Runway 发布 Runway Dev MCP 托管服务器，让编码 Agent 自动完成集成

您的编码代理现在可以规划、配置和调试您的Runway Dev集成。

今天，我们推出了Runway Dev MCP，这是一个托管的MCP服务器，将你的Runway Dev：https：//dev.runway.com/账户连接到你已有的编码工具。Runway Dev将Runway及其他实验室的前沿模型与创意工具结合起来，帮助你从创意到制作质量的媒体。现在，编写你集成的代理可以选择合适的模型，配置运行的工具，并调试，并以更高的代币效率进行调试。

Runway Dev MCP 赋予你的编码代理对集成的完全自主权，从开发中的第一个请求到生产环境中的实时调试。

在编写单个API调用之前，你的代理可以使用Runway Dev MCP研究项目可以使用哪些模型，以及价格、能力和每个模型接受的输入。然后它会拉取所选模型的精确请求模式，这样它就能在第一次尝试时正确调用模型，而不是猜测。

你的客服可以做得比单一型号更进一步。借助Runway Dev MCP，它可以创建并配置一个模型路由器：https：//runway.com/news/company-news/introducing-runway-media-router，优化多个型号的成本、延迟或质量，然后建议哪些型号符合条件，以及每代设定的成本上限。每次提到路由器的通话都会与这些规则匹配，然后根据Runway自身的质量、价格和速度数据进行排名。你的客服还可以查询路由器为任何通话选择了哪个型号。

字符也是一样：https：//runway.com/product/characters。你的代理创建一个字符，列出并检查项目已有的字符，并附加或分离背后的知识文档。

当生成失败时，你的代理会用Runway Dev MCP查找任务，而不是手动写检查。它调用定义的工具，回读具体原因：审核拒绝、资产大小限制、请求体格式错误。然后修复请求并再次运行，甚至在你注意到失败之前。

Runway Dev 是为所有通过编码代理推动集成并拥有它的人打造的：工程师、产品经理和独立创始人。我们设计它让这些事情不必在代理之外发生。

Runway Dev MCP 是其中的一部分。文档是另一部分：它是为编程代理使用作为上下文而构建的，因此代理可以获取其所需的整个 API 范围。

快速入门菜单是第三部分，它的设计目标是尽可能减少摩擦：创建密钥后，它会打开 Claude Code、Codex 或 Cursor，并自动在消息中输入内容供您的代理使用，以便代理可以接管，安装 SDK 并发起第一次调用。

设置只需几秒钟。在 Claude、ChatGPT、Codex 或 Cursor 中，将 Runway Dev MCP 添加为自定义连接器，网址为 https://dev.runwayml.com/mcp，并在提示时批准访问权限。

Runway Dev 是我们的生成媒体 API 平台。可访问来自 Runway 及其他前沿实验室的最佳模型，以及用于以最小努力大规模获取一流结果的工具。我们的专业知识已内置于平台中，因此模型评估、选择、提示和链式操作变得轻而易举。

其设计符合企业级标准：SOC 2 二型合规、知识产权保障、内置内容审核与 99.9% 正常运行时间。每个模型供应商都通过 Runway 的供应商管理程序，我们为平台上的每个模型协商无训练限制和数据保留条款。支出管理和每个模型的成本可见性是内置的。企业客户：https://runway.com/enterprise 还可获得专属客户经理和 Runway 技术艺术家的支持。

Your coding agent can now plan, configure and debug your Runway Dev integration.

Today, we're launching Runway Dev MCP, a hosted MCP server that connects your Runway Dev：https://dev.runway.com/ account to the coding tools you already work in. Runway Dev pairs frontier models from Runway and other labs with the creative tooling that takes you from idea to production-quality media. Now, the agent that wrote your integration can pick the right model for it, configure the tools it runs on and debug it, with better token efficiency.

Runway Dev MCP gives your coding agent full autonomy over the integration, from the first request in development to live debugging in production.

Before writing a single API call, your agent can use the Runway Dev MCP to research which models a project can use, along with prices, capabilities and the inputs each one accepts. It then pulls the exact request schema for the model it picks, so it calls the model correctly on the first try instead of guessing.

Your agent can go further than a single model. With the Runway Dev MCP, it can create and configure a Model Router：https://runway.com/news/company-news/introducing-runway-media-router that optimizes for cost, latency or quality across several models, then suggest which models to make eligible and what cost limit to set per generation. Every call that references the router is matched against those rules, then ranked on Runway's own quality, price and speed data. Your agent can also look up which model the router picked for any given call.

The same goes for Characters：https://runway.com/product/characters. Your agent creates one, lists and inspects the ones a project already has, and attaches or detaches the knowledge documents behind them.

When a generation fails, your agent uses the Runway Dev MCP to look up the task instead of writing that check by hand. It calls a defined tool and reads back the specific reason: a moderation rejection, an asset size limit, a malformed request body. Then it fixes the request and runs it again, before you'd even notice it failed.

Runway Dev is built for anyone who drives their integration through a coding agent, then owns it: engineers, product managers and solo founders. We designed it so nothing about that has to happen outside the agent.

Runway Dev MCP is one piece of that. Documentation is another: it's built for coding agents to consume as context, so an agent can take in the entire API surface it needs.

The Quickstart menu is the third piece, and it's built for the least friction possible: create the key, and it opens Claude Code, Codex or Cursor with a message for your agent already typed in, so it can take over, installing the SDK and making the first call.

Setting up takes seconds. In Claude, ChatGPT, Codex or Cursor, add Runway Dev MCP as a custom connector at https://dev.runwayml.com/mcp and approve access when prompted.

Runway Dev is our API platform for generative media. Access the best models from Runway and other frontier labs, plus the tools to get best-in-class output with the least effort, at scale. Our expertise is built into the platform, so model evaluation, selection, prompting and chaining are the easy part.

It's enterprise-grade by design: SOC 2 Type II compliance, IP indemnification, built-in content moderation and 99.9% uptime. Every model provider passes Runway's vendor management program, and we negotiate no-training restrictions and data-retention terms for every model on the platform. Spend management and per-model cost visibility are built in. Enterprise customers ：https://runway.com/enterprisealso get a dedicated account manager and support from Runway's technical artists.