Fal.ai MCP Server

Instructions

Implementation Reference

• src/index.ts:286-322 (handler)

  Handler function for 'run_model' tool: parses input with RunModelSchema, then either streams or subscribes to the specified Fal.ai model using fal.stream or fal.subscribe, returning JSON results.

  case "run_model": {
    const params = RunModelSchema.parse(args);
    
    if (params.stream) {
      const stream = await fal.stream(params.model_id, {
        input: params.input,
      });
      
      const events: any[] = [];
      for await (const event of stream) {
        events.push(event);
      }
      
      return {
        content: [
          {
            type: "text",
            text: JSON.stringify(events, null, 2),
          },
        ],
      };
    } else {
      const result = await fal.subscribe(params.model_id, {
        input: params.input,
        logs: true,
      });
      
      return {
        content: [
          {
            type: "text",
            text: JSON.stringify(result, null, 2),
          },
        ],
      };
    }
  }

• src/index.ts:71-75 (schema)

  Zod schema defining input parameters for the 'run_model' tool: model_id, input, and optional stream flag.

  const RunModelSchema = z.object({
    model_id: z.string().describe("The Fal.ai model endpoint ID (e.g., 'fal-ai/flux/dev')"),
    input: z.record(z.any()).describe("Input parameters for the model"),
    stream: z.boolean().default(false).optional().describe("Whether to stream results"),
  });

• src/index.ts:138-160 (registration)

  Tool registration in the list of tools returned by ListToolsRequestHandler, including name, description, and inputSchema matching the Zod schema.

  {
    name: "run_model",
    description: "Run any Fal.ai model with custom parameters",
    inputSchema: {
      type: "object",
      properties: {
        model_id: {
          type: "string",
          description: "The Fal.ai model endpoint ID",
        },
        input: {
          type: "object",
          description: "Input parameters for the model",
        },
        stream: {
          type: "boolean",
          description: "Whether to stream results",
          default: false,
        },
      },
      required: ["model_id", "input"],
    },
  },