replicate-flux-mcp

Instructions

Implementation Reference

• src/tools/createPrediction.ts:7-30 (handler)

  The handler function that executes the create_prediction tool: creates a prediction via Replicate API with the given input, polls for completion, and returns the result as text JSON.

  export const registerCreatePredictionTool = async (
    input: CreatePredictionParams
  ): Promise<CallToolResult> => {
    try {
      const prediction = await replicate.predictions.create({
        model: CONFIG.imageModelId,
        input,
      });
  
      await replicate.predictions.get(prediction.id);
      const completed = await pollForCompletion(prediction.id);
  
      return {
        content: [
          {
            type: "text",
            text: JSON.stringify(completed || "Processing timed out", null, 2),
          },
        ],
      };
    } catch (error) {
      handleError(error);
    }
  };

• src/types/index.ts:3-69 (schema)

  Input schema using Zod for validating parameters of the create_prediction tool, including prompt, seed, aspect ratio, etc.

  export const createPredictionSchema = {
    prompt: z.string().min(1).describe("Prompt for generated image"),
    seed: z
      .number()
      .int()
      .optional()
      .describe("Random seed. Set for reproducible generation"),
    go_fast: z
      .boolean()
      .default(true)
      .describe(
        "Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16"
      ),
    megapixels: z
      .enum(["1", "0.25"])
      .default("1")
      .describe("Approximate number of megapixels for generated image"),
    num_outputs: z
      .number()
      .int()
      .min(1)
      .max(4)
      .default(1)
      .describe("Number of outputs to generate"),
    aspect_ratio: z
      .enum([
        "1:1",
        "16:9",
        "21:9",
        "3:2",
        "2:3",
        "4:5",
        "5:4",
        "3:4",
        "4:3",
        "9:16",
        "9:21",
      ])
      .default("1:1")
      .describe("Aspect ratio for the generated image"),
    output_format: z
      .enum(["webp", "jpg", "png"])
      .default("webp")
      .describe("Format of the output images"),
    output_quality: z
      .number()
      .int()
      .min(0)
      .max(100)
      .default(80)
      .describe(
        "Quality when saving the output images, from 0 to 100. 100 is best quality, 0 is lowest quality. Not relevant for .png outputs"
      ),
    num_inference_steps: z
      .number()
      .int()
      .min(1)
      .max(4)
      .default(4)
      .describe(
        "Number of denoising steps. 4 is recommended, and lower number of steps produce lower quality outputs, faster."
      ),
    disable_safety_checker: z
      .boolean()
      .default(false)
      .describe("Disable safety checker for generated images."),
  };

• src/tools/index.ts:51-56 (registration)

  Registration of the 'create_prediction' tool on the MCP server, providing name, description, schema, and handler function.

  server.tool(
    "create_prediction",
    "Generate an prediction from a text prompt using Flux Schnell model",
    createPredictionSchema,
    registerCreatePredictionTool
  );