FAL Image/Video MCP Server

Instructions

Implementation Reference

• src/index.ts:566-625 (handler)

  Handler function that executes the 'ltx_video' tool by calling the FAL.ai endpoint 'fal-ai/ltx-video-13b-distilled/image-to-video', processing input parameters, handling video output including downloads and data URLs.
  private async handleImageToVideo(args: any, model: any) { const { image_url, prompt, duration = '5', aspect_ratio = '16:9', negative_prompt, cfg_scale } = args; try { // Configure FAL client lazily with query config override configureFalClient(this.currentQueryConfig); const inputParams: any = { image_url, prompt }; // Add optional parameters if (duration) inputParams.duration = duration; if (aspect_ratio) inputParams.aspect_ratio = aspect_ratio; if (negative_prompt) inputParams.negative_prompt = negative_prompt; if (cfg_scale !== undefined) inputParams.cfg_scale = cfg_scale; const result = await fal.subscribe(model.endpoint, { input: inputParams }); const videoData = result.data as FalVideoResult; const videoProcessed = await downloadAndProcessVideo(videoData.video.url, model.id); return { content: [ { type: 'text', text: JSON.stringify({ model: model.name, id: model.id, endpoint: model.endpoint, input_image: image_url, prompt, video: { url: videoData.video.url, localPath: videoProcessed.localPath, ...(videoProcessed.dataUrl && { dataUrl: videoProcessed.dataUrl }), width: videoData.video.width, height: videoData.video.height, }, metadata: inputParams, download_path: DOWNLOAD_PATH, data_url_settings: { enabled: ENABLE_DATA_URLS, max_size_mb: Math.round(MAX_DATA_URL_SIZE / 1024 / 1024), }, autoopen_settings: { enabled: AUTOOPEN, note: AUTOOPEN ? "Files automatically opened with default application" : "Auto-open disabled" }, }, null, 2), }, ], }; } catch (error) { throw new Error(`${model.name} generation failed: ${error}`); } }
• src/index.ts:380-390 (schema)

  Input schema definition for image-to-video tools, including ltx_video, specifying required image_url and prompt, and optional parameters like duration, aspect_ratio, etc.
  } else if (category === 'imageToVideo') { baseSchema.inputSchema.properties = { image_url: { type: 'string', description: 'URL of the input image' }, prompt: { type: 'string', description: 'Motion description prompt' }, duration: { type: 'string', enum: ['5', '10'], default: '5', description: 'Video duration in seconds' }, aspect_ratio: { type: 'string', enum: ['16:9', '9:16', '1:1'], default: '16:9' }, negative_prompt: { type: 'string', description: 'What to avoid in the video' }, cfg_scale: { type: 'number', default: 0.5, minimum: 0, maximum: 1, description: 'How closely to follow the prompt' } }; baseSchema.inputSchema.required = ['image_url', 'prompt']; }
• src/index.ts:120-120 (registration)

  Registration of the ltx_video tool in the MODEL_REGISTRY.imageToVideo array, defining its ID, endpoint, name, and description.
  { id: 'ltx_video', endpoint: 'fal-ai/ltx-video-13b-distilled/image-to-video', name: 'LTX Video', description: 'Fast and high-quality image-to-video conversion' },
• src/index.ts:406-408 (registration)

  Dynamic registration of ltx_video tool schema during tools/list response by iterating over imageToVideo models.
  for (const model of MODEL_REGISTRY.imageToVideo) { tools.push(this.generateToolSchema(model, 'imageToVideo')); }
• src/index.ts:480-482 (handler)

  Dispatch logic in CallToolRequestSchema handler that routes 'ltx_video' calls to the handleImageToVideo function based on model ID.
  } else if (MODEL_REGISTRY.imageToVideo.find(m => m.id === name)) { return await this.handleImageToVideo(args, model); }