Runway API MCP Server

Instructions

Implementation Reference

• src/index.ts:114-137 (handler)

  The asynchronous handler function for the runway_generateImage tool. It invokes the Runway /text_to_image API endpoint asynchronously, waits for task completion, and returns a success message with the image URL or the task details.

  async ({ promptText, ratio, referenceImages }) => {
    const task = await callRunwayAsync("/text_to_image", {
      method: "POST",
      body: JSON.stringify({
        model: "gen4_image",
        promptText,
        ratio,
        referenceImages,
      }),
    });
  
    if (task.status === "SUCCEEDED") {
      return {
        content: [
          {
            type: "text",
            text: `Here is the URL of the image: ${task.output[0]}. Return to the user, as a markdown link, the URL of the image and the prompt that was used to generate the image.`,
          },
        ],
      };
    } else {
      return { content: [{ type: "text", text: JSON.stringify(task) }] };
    }
  }

• src/index.ts:107-113 (schema)

  Zod input schema for the tool: required promptText and ratio (strings), optional referenceImages (array of {uri: string, tag?: string}).

  {
    promptText: z.string(),
    ratio: z.string(),
    referenceImages: z
      .array(z.object({ uri: z.string(), tag: z.string().optional() }))
      .optional(),
  },

• src/index.ts:100-138 (registration)

  Full registration of the 'runway_generateImage' tool via server.tool(), including the tool name, multi-line description, input schema, and inline handler function.

  server.tool(
    "runway_generateImage",
    `Generate an image from a text prompt and optional reference images. Available ratios are 1920:1080, 1080:1920, 1024:1024, 
    1360:768, 1080:1080, 1168:880, 1440:1080, 1080:1440, 1808:768, 2112:912, 1280:720, 720:1280, 
    720:720, 960:720, 720:960, 1680:720. Use 1920:1080 by default. It also accepts reference images, in the form of either a url or a base64 encoded image. 
    Each reference image has a tag, which is a string that refers to the image from the user prompt. For example, if the user prompt is "IMG_1 on a red background", 
    and the reference image has the tag "IMG_1", the model will use that reference image to generate the image. The return of this function will contain a url to the generated image.`,
    {
      promptText: z.string(),
      ratio: z.string(),
      referenceImages: z
        .array(z.object({ uri: z.string(), tag: z.string().optional() }))
        .optional(),
    },
    async ({ promptText, ratio, referenceImages }) => {
      const task = await callRunwayAsync("/text_to_image", {
        method: "POST",
        body: JSON.stringify({
          model: "gen4_image",
          promptText,
          ratio,
          referenceImages,
        }),
      });
  
      if (task.status === "SUCCEEDED") {
        return {
          content: [
            {
              type: "text",
              text: `Here is the URL of the image: ${task.output[0]}. Return to the user, as a markdown link, the URL of the image and the prompt that was used to generate the image.`,
            },
          ],
        };
      } else {
        return { content: [{ type: "text", text: JSON.stringify(task) }] };
      }
    }
  );