@mixio-pro/kalaasetu-mcp 1.0.4 → 1.0.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mixio-pro/kalaasetu-mcp",
-  "version": "1.0.4",
+  "version": "1.0.6",
   "description": "A powerful Model Context Protocol server providing AI tools for content generation and analysis",
   "type": "module",
   "module": "src/index.ts",

package/src/storage/index.ts

@@ -0,0 +1,27 @@
+import type { StorageProvider } from "./interface";
+import { LocalStorageProvider } from "./local";
+import { PayloadStorageProvider } from "./payload";
+
+let storageInstance: StorageProvider | null = null;
+
+export function getStorage(): StorageProvider {
+  if (!storageInstance) {
+    const type = process.env.STORAGE_PROVIDER || "local";
+    console.error(`Initializing storage provider: ${type}`); // Log to stderr for debug
+
+    if (type === "payload") {
+      storageInstance = new PayloadStorageProvider(
+        process.env.PAYLOAD_API_URL || "http://localhost:3000",
+        process.env.PAYLOAD_API_KEY || ""
+      );
+    } else {
+      storageInstance = new LocalStorageProvider(process.cwd());
+    }
+
+    // Initialize async (fire and forget or handle properly in app startup)
+    storageInstance
+      .init()
+      .catch((err) => console.error("Failed to init storage:", err));
+  }
+  return storageInstance;
+}

package/src/storage/interface.ts

@@ -0,0 +1,7 @@
+export interface StorageProvider {
+  init(): Promise<void>;
+  readFile(path: string): Promise<Buffer>;
+  writeFile(path: string, data: Buffer | string): Promise<string>; // Returns path or URL
+  exists(path: string): Promise<boolean>;
+  getPublicUrl(path: string): Promise<string>;
+}

package/src/storage/local.ts

@@ -0,0 +1,55 @@
+import * as fs from "fs";
+import * as path from "path";
+import type { StorageProvider } from "./interface";
+
+export class LocalStorageProvider implements StorageProvider {
+  private basePath: string;
+
+  constructor(basePath: string = process.cwd()) {
+    this.basePath = basePath;
+  }
+
+  async init(): Promise<void> {
+    // No-op for local
+  }
+
+  async readFile(filePath: string): Promise<Buffer> {
+    // Handle absolute paths by checking if it starts with basePath or just use it if it exists
+    let fullPath = filePath;
+    if (!path.isAbsolute(filePath)) {
+      fullPath = path.resolve(this.basePath, filePath);
+    }
+    return fs.promises.readFile(fullPath);
+  }
+
+  async writeFile(filePath: string, data: Buffer | string): Promise<string> {
+    let fullPath = filePath;
+    if (!path.isAbsolute(filePath)) {
+      fullPath = path.resolve(this.basePath, filePath);
+    }
+
+    const dir = path.dirname(fullPath);
+    if (!fs.existsSync(dir)) {
+      await fs.promises.mkdir(dir, { recursive: true });
+    }
+    await fs.promises.writeFile(fullPath, data);
+    return fullPath;
+  }
+
+  async exists(filePath: string): Promise<boolean> {
+    let fullPath = filePath;
+    if (!path.isAbsolute(filePath)) {
+      fullPath = path.resolve(this.basePath, filePath);
+    }
+    return fs.existsSync(fullPath);
+  }
+
+  async getPublicUrl(filePath: string): Promise<string> {
+    // For local, we just return the absolute path
+    let fullPath = filePath;
+    if (!path.isAbsolute(filePath)) {
+      fullPath = path.resolve(this.basePath, filePath);
+    }
+    return fullPath;
+  }
+}

package/src/storage/payload.ts

@@ -0,0 +1,46 @@
+import type { StorageProvider } from "./interface";
+
+export class PayloadStorageProvider implements StorageProvider {
+  private apiUrl: string;
+  private apiKey: string;
+  private collection: string;
+
+  constructor(apiUrl: string, apiKey: string, collection: string = "media") {
+    this.apiUrl = apiUrl;
+    this.apiKey = apiKey;
+    this.collection = collection;
+  }
+
+  async init(): Promise<void> {
+    console.log("Initializing Payload Storage Provider...");
+    // TODO: Verify connection to Payload CMS
+  }
+
+  async readFile(filePath: string): Promise<Buffer> {
+    // TODO: Implement fetching file from Payload CMS
+    // 1. Search for file by filename or ID
+    // 2. Download the file buffer
+    console.log(`[Payload] Reading file: ${filePath}`);
+    throw new Error("PayloadStorageProvider.readFile not implemented yet.");
+  }
+
+  async writeFile(filePath: string, data: Buffer | string): Promise<string> {
+    // TODO: Implement uploading file to Payload CMS
+    // 1. Create FormData
+    // 2. POST to /api/{collection}
+    console.log(`[Payload] Writing file: ${filePath}`);
+    throw new Error("PayloadStorageProvider.writeFile not implemented yet.");
+  }
+
+  async exists(filePath: string): Promise<boolean> {
+    // TODO: Check if file exists in Payload
+    console.log(`[Payload] Checking existence: ${filePath}`);
+    return false;
+  }
+
+  async getPublicUrl(filePath: string): Promise<string> {
+    // TODO: Return the public URL of the file in Payload
+    console.log(`[Payload] Getting public URL: ${filePath}`);
+    return `${this.apiUrl}/${this.collection}/${filePath}`;
+  }
+}

package/src/tools/gemini.ts

@@ -1,17 +1,26 @@
 import { z } from "zod";
-import { GoogleGenAI, createPartFromUri, createUserContent } from "@google/genai";
+import {
+  GoogleGenAI,
+  createPartFromUri,
+  createUserContent,
+} from "@google/genai";
 import * as fs from "fs";
+import * as path from "path";
+import * as os from "os";
 import * as wav from "wav";
+import { PassThrough } from "stream";
+import { getStorage } from "../storage";
 
 const ai = new GoogleGenAI({
   apiKey: process.env.GEMINI_API_KEY || "",
 });
 
-function fileToGenerativePart(filePath: string) {
-  if (!fs.existsSync(filePath)) {
+async function fileToGenerativePart(filePath: string) {
+  const storage = getStorage();
+  if (!(await storage.exists(filePath))) {
     throw new Error(`File not found: ${filePath}`);
   }
-  const imageBytes = fs.readFileSync(filePath);
+  const imageBytes = await storage.readFile(filePath);
   return {
     inlineData: {
       data: Buffer.from(imageBytes).toString("base64"),
@@ -21,22 +30,37 @@ function fileToGenerativePart(filePath: string) {
 }
 
 // Helper function to save WAV file
-function saveWaveFile(
+// Helper function to save WAV file
+async function saveWaveFile(
   filename: string,
   pcmData: Buffer,
   channels = 1,
   rate = 24000,
-  sampleWidth = 2,
+  sampleWidth = 2
 ): Promise<void> {
   return new Promise((resolve, reject) => {
-    const writer = new wav.FileWriter(filename, {
+    const writer = new wav.Writer({
       channels,
       sampleRate: rate,
       bitDepth: sampleWidth * 8,
     });
 
-    writer.on('finish', resolve);
-    writer.on('error', reject);
+    const stream = new PassThrough();
+    const chunks: Buffer[] = [];
+
+    writer.pipe(stream);
+    stream.on("data", (chunk) => chunks.push(chunk));
+    stream.on("end", async () => {
+      try {
+        const wavBuffer = Buffer.concat(chunks);
+        const storage = getStorage();
+        await storage.writeFile(filename, wavBuffer);
+        resolve();
+      } catch (err) {
+        reject(err);
+      }
+    });
+    writer.on("error", reject);
 
     writer.write(pcmData);
     writer.end();
@@ -45,31 +69,59 @@ function saveWaveFile(
 
 // Helper function to check if URL is YouTube URL
 function isYouTubeUrl(url: string): boolean {
-  return url.includes('youtube.com/watch') || url.includes('youtu.be');
+  return url.includes("youtube.com/watch") || url.includes("youtu.be");
 }
 
 // Helper function to get file size in bytes
-function getFileSize(filePath: string): number {
-  const stats = fs.statSync(filePath);
-  return stats.size;
+async function getFileSize(filePath: string): Promise<number> {
+  const storage = getStorage();
+  const buffer = await storage.readFile(filePath);
+  return buffer.length;
 }
 
+// Helper function to upload file to Gemini API
 // Helper function to upload file to Gemini API
 async function uploadFileToGemini(filePath: string): Promise<any> {
   try {
+    const storage = getStorage();
+    // For Gemini API, we need a local file path.
+    // If storage is local, we can use the path directly (if we can resolve it).
+    // If storage is remote, we must download to a temp file.
+
+    let localPath = filePath;
+    let isTemp = false;
+
+    // Check if we can get a local path from storage (hacky check for LocalStorageProvider)
+    // A better way is to always download to temp if not sure, or ask storage for a local path.
+    // For now, let's assume we need to download if it's not a local file system path that exists.
+
+    if (!fs.existsSync(filePath)) {
+      // Try to read from storage and write to temp
+      const buffer = await storage.readFile(filePath);
+      const tempDir = os.tmpdir();
+      const tempFilePath = path.join(tempDir, path.basename(filePath));
+      fs.writeFileSync(tempFilePath, buffer);
+      localPath = tempFilePath;
+      isTemp = true;
+    }
+
     const uploadedFile = await ai.files.upload({
-      file: filePath,
+      file: localPath,
     });
-    
+
+    if (isTemp) {
+      fs.unlinkSync(localPath);
+    }
+
     // Wait for file processing to complete
     let getFile = await ai.files.get({ name: uploadedFile.name! });
-    while (getFile.state === 'PROCESSING') {
-      await new Promise(resolve => setTimeout(resolve, 3000));
+    while (getFile.state === "PROCESSING") {
+      await new Promise((resolve) => setTimeout(resolve, 3000));
       getFile = await ai.files.get({ name: uploadedFile.name! });
     }
 
-    if (getFile.state === 'FAILED') {
-      throw new Error('File processing failed');
+    if (getFile.state === "FAILED") {
+      throw new Error("File processing failed");
     }
 
     return getFile;
@@ -79,41 +131,58 @@ async function uploadFileToGemini(filePath: string): Promise<any> {
 }
 
 // Helper function to process video input intelligently
-async function processVideoInput(input: string, config?: { fps?: number; startOffset?: string; endOffset?: string }): Promise<any> {
+async function processVideoInput(
+  input: string,
+  config?: { fps?: number; startOffset?: string; endOffset?: string }
+): Promise<any> {
   if (isYouTubeUrl(input)) {
     return {
       fileData: {
         fileUri: input,
-        mimeType: 'video/*',
-        videoMetadata: config ? {
-          fps: config.fps,
-          startOffset: config.startOffset,
-          endOffset: config.endOffset
-        } : undefined
-      }
+        mimeType: "video/*",
+        videoMetadata: config
+          ? {
+              fps: config.fps,
+              startOffset: config.startOffset,
+              endOffset: config.endOffset,
+            }
+          : undefined,
+      },
     };
   } else {
     // Local file processing - use File Upload API
-    if (!fs.existsSync(input)) {
+    const storage = getStorage();
+    if (!(await storage.exists(input))) {
       throw new Error(`Video file not found: ${input}`);
     }
-    
+
     // Upload file to Gemini API
     const uploadedFile = await uploadFileToGemini(input);
-    
+
     return uploadedFile;
   }
 }
 
 export const geminiTextToImage = {
   name: "geminiTextToImage",
-  description: "Generate images from text prompts using Gemini 2.5 Flash Image model",
+  description:
+    "Generate images from text prompts using Gemini 2.5 Flash Image model",
   parameters: z.object({
     prompt: z.string().describe("Text description of the image to generate"),
-    aspect_ratio: z.string().optional().describe("Aspect ratio: 1:1, 3:4, 4:3, 9:16, or 16:9"),
-    output_path: z.string().optional().describe("File path to save the generated image"),
+    aspect_ratio: z
+      .string()
+      .optional()
+      .describe("Aspect ratio: 1:1, 3:4, 4:3, 9:16, or 16:9"),
+    output_path: z
+      .string()
+      .optional()
+      .describe("File path to save the generated image"),
   }),
-  execute: async (args: { prompt: string; aspect_ratio?: string; output_path?: string }) => {
+  execute: async (args: {
+    prompt: string;
+    aspect_ratio?: string;
+    output_path?: string;
+  }) => {
     try {
       const response = await ai.models.generateContent({
         model: "gemini-2.5-flash-image",
@@ -133,10 +202,17 @@ export const geminiTextToImage = {
           } else if (part.inlineData?.data) {
             const imageData = part.inlineData.data;
             if (args.output_path) {
-              fs.writeFileSync(args.output_path, Buffer.from(imageData, "base64"));
+              const storage = getStorage();
+              await storage.writeFile(
+                args.output_path,
+                Buffer.from(imageData, "base64")
+              );
               result += `\nImage saved to: ${args.output_path}`;
             } else {
-              result += `\nGenerated image (base64): ${imageData.substring(0, 100)}...`;
+              result += `\nGenerated image (base64): ${imageData.substring(
+                0,
+                100
+              )}...`;
             }
           }
         }
@@ -150,20 +226,33 @@ export const geminiTextToImage = {
 
 export const geminiEditImage = {
   name: "geminiEditImage",
-  description: "Edit existing images with text instructions using Gemini 2.5 Flash Image Preview",
+  description:
+    "Edit existing images with text instructions using Gemini 2.5 Flash Image Preview",
   parameters: z.object({
     image_path: z.string().describe("Path to the source image file"),
     prompt: z.string().describe("Text instructions for editing the image"),
-    output_path: z.string().optional().describe("File path to save the edited image"),
-    reference_images: z.array(z.string()).optional().describe("Additional image paths for reference"),
+    output_path: z
+      .string()
+      .optional()
+      .describe("File path to save the edited image"),
+    reference_images: z
+      .array(z.string())
+      .optional()
+      .describe("Additional image paths for reference"),
   }),
-  execute: async (args: { image_path: string; prompt: string; output_path?: string; reference_images?: string[] }) => {
+  execute: async (args: {
+    image_path: string;
+    prompt: string;
+    output_path?: string;
+    reference_images?: string[];
+  }) => {
     try {
-      const contents: any[] = [args.prompt, fileToGenerativePart(args.image_path)];
-      
+      const imagePart = await fileToGenerativePart(args.image_path);
+      const contents: any[] = [args.prompt, imagePart];
+
       if (args.reference_images) {
         for (const refPath of args.reference_images) {
-          contents.push(fileToGenerativePart(refPath));
+          contents.push(await fileToGenerativePart(refPath));
         }
       }
 
@@ -180,10 +269,17 @@ export const geminiEditImage = {
           } else if (part.inlineData?.data) {
             const imageData = part.inlineData.data;
             if (args.output_path) {
-              fs.writeFileSync(args.output_path, Buffer.from(imageData, "base64"));
+              const storage = getStorage();
+              await storage.writeFile(
+                args.output_path,
+                Buffer.from(imageData, "base64")
+              );
               result += `\nEdited image saved to: ${args.output_path}`;
             } else {
-              result += `\nEdited image (base64): ${imageData.substring(0, 100)}...`;
+              result += `\nEdited image (base64): ${imageData.substring(
+                0,
+                100
+              )}...`;
             }
           }
         }
@@ -197,9 +293,12 @@ export const geminiEditImage = {
 
 export const geminiAnalyzeImages = {
   name: "geminiAnalyzeImages",
-  description: "Analyze and describe images using Gemini 2.5 Pro with advanced multimodal understanding",
+  description:
+    "Analyze and describe images using Gemini 2.5 Pro with advanced multimodal understanding",
   parameters: z.object({
-    image_paths: z.array(z.string()).describe("Array of image file paths to analyze"),
+    image_paths: z
+      .array(z.string())
+      .describe("Array of image file paths to analyze"),
     prompt: z.string().describe("Text prompt or question about the images"),
   }),
   execute: async (args: { image_paths: string[]; prompt: string }) => {
@@ -208,12 +307,12 @@ export const geminiAnalyzeImages = {
       if (!args.image_paths) {
         throw new Error("Image paths not provided");
       }
-      
+
       // Convert to array if passed as string
       let imagePaths: string[];
-      if (typeof args.image_paths === 'string') {
+      if (typeof args.image_paths === "string") {
         const strValue = args.image_paths as string;
-        if (strValue.startsWith('[') && strValue.endsWith(']')) {
+        if (strValue.startsWith("[") && strValue.endsWith("]")) {
           try {
             imagePaths = JSON.parse(strValue);
           } catch {
@@ -227,15 +326,15 @@ export const geminiAnalyzeImages = {
       } else {
         throw new Error("Invalid image_paths: must be array or string");
       }
-      
+
       if (imagePaths.length === 0) {
         throw new Error("At least one image path must be provided");
       }
-      
+
       const contents: any[] = [args.prompt];
-      
+
       for (const imagePath of imagePaths) {
-        contents.push(fileToGenerativePart(imagePath));
+        contents.push(await fileToGenerativePart(imagePath));
       }
 
       const response = await ai.models.generateContent({
@@ -260,41 +359,56 @@ export const geminiAnalyzeImages = {
 
 export const geminiSingleSpeakerTts = {
   name: "geminiSingleSpeakerTts",
-  description: "Generate single speaker voice audio from text using Gemini 2.5 Pro Preview TTS model",
+  description:
+    "Generate single speaker voice audio from text using Gemini 2.5 Pro Preview TTS model",
   parameters: z.object({
     text: z.string().describe("Text to convert to speech"),
-    voice_name: z.string().describe("Voice name from supported options. Use Kore, Erinome or Despina for the female voices and Enceladus for male."),
-    output_path: z.string().optional().describe("Output WAV file path (optional, defaults to timestamp-based filename)"),
+    voice_name: z
+      .string()
+      .describe(
+        "Voice name from supported options. Use Kore, Erinome or Despina for the female voices and Enceladus for male."
+      ),
+    output_path: z
+      .string()
+      .optional()
+      .describe(
+        "Output WAV file path (optional, defaults to timestamp-based filename)"
+      ),
   }),
-  execute: async (args: { text: string; voice_name: string; output_path?: string }) => {
+  execute: async (args: {
+    text: string;
+    voice_name: string;
+    output_path?: string;
+  }) => {
     try {
       const response = await ai.models.generateContent({
         model: "gemini-2.5-pro-preview-tts",
         contents: [{ parts: [{ text: args.text }] }],
         config: {
-          responseModalities: ['AUDIO'],
+          responseModalities: ["AUDIO"],
           speechConfig: {
             voiceConfig: {
-              prebuiltVoiceConfig: { 
-                voiceName: args.voice_name || 'Despina'
+              prebuiltVoiceConfig: {
+                voiceName: args.voice_name || "Despina",
               },
             },
           },
         },
       });
 
-      const data = response.candidates?.[0]?.content?.parts?.[0]?.inlineData?.data;
+      const data =
+        response.candidates?.[0]?.content?.parts?.[0]?.inlineData?.data;
       if (!data) {
         throw new Error("No audio data received from Gemini API");
       }
 
-      const audioBuffer = Buffer.from(data, 'base64');
-      
+      const audioBuffer = Buffer.from(data, "base64");
+
       // Generate output filename if not provided
       const outputPath = args.output_path || `voice_output_${Date.now()}.wav`;
-      
+
       await saveWaveFile(outputPath, audioBuffer);
-      
+
       return `Audio generated successfully: ${outputPath}`;
     } catch (error: any) {
       throw new Error(`Voice generation failed: ${error.message}`);
@@ -304,27 +418,59 @@ export const geminiSingleSpeakerTts = {
 
 export const geminiAnalyzeVideos = {
   name: "geminiAnalyzeVideos",
-  description: "Analyze and understand video content using Gemini 2.5 Flash model. Intelligently handles YouTube URLs and local videos (files <20MB processed inline, ≥20MB uploaded via File API). Supports timestamp queries, clipping, and custom frame rates with default 5 FPS for local videos to optimize processing.",
+  description:
+    "Analyze and understand video content using Gemini 2.5 Flash model. Intelligently handles YouTube URLs and local videos (files <20MB processed inline, ≥20MB uploaded via File API). Supports timestamp queries, clipping, and custom frame rates with default 5 FPS for local videos to optimize processing.",
   parameters: z.object({
-    video_inputs: z.array(z.string()).describe("Array of video inputs - mix of local file paths and YouTube URLs (max 10 videos). Local files <20MB processed inline, larger files uploaded via File API automatically."),
-    prompt: z.string().describe("Text prompt or question about the videos. Use MM:SS format for timestamp references (e.g., 'What happens at 01:30?')."),
-    fps: z.number().optional().describe("Frame rate for video processing (default: 5 FPS for local videos to reduce file size, 1 FPS for YouTube URLs)"),
-    start_offset: z.string().optional().describe("Clip start time in seconds with 's' suffix (e.g., '40s')"),
-    end_offset: z.string().optional().describe("Clip end time in seconds with 's' suffix (e.g., '80s')"),
-    media_resolution: z.string().optional().describe("Media resolution: 'default' or 'low' (low resolution uses ~100 tokens/sec vs 300 tokens/sec)"),
+    video_inputs: z
+      .array(z.string())
+      .describe(
+        "Array of video inputs - mix of local file paths and YouTube URLs (max 10 videos). Local files <20MB processed inline, larger files uploaded via File API automatically."
+      ),
+    prompt: z
+      .string()
+      .describe(
+        "Text prompt or question about the videos. Use MM:SS format for timestamp references (e.g., 'What happens at 01:30?')."
+      ),
+    fps: z
+      .number()
+      .optional()
+      .describe(
+        "Frame rate for video processing (default: 5 FPS for local videos to reduce file size, 1 FPS for YouTube URLs)"
+      ),
+    start_offset: z
+      .string()
+      .optional()
+      .describe("Clip start time in seconds with 's' suffix (e.g., '40s')"),
+    end_offset: z
+      .string()
+      .optional()
+      .describe("Clip end time in seconds with 's' suffix (e.g., '80s')"),
+    media_resolution: z
+      .string()
+      .optional()
+      .describe(
+        "Media resolution: 'default' or 'low' (low resolution uses ~100 tokens/sec vs 300 tokens/sec)"
+      ),
   }),
-  execute: async (args: { video_inputs: string[]; prompt: string; fps?: number; start_offset?: string; end_offset?: string; media_resolution?: string }) => {
+  execute: async (args: {
+    video_inputs: string[];
+    prompt: string;
+    fps?: number;
+    start_offset?: string;
+    end_offset?: string;
+    media_resolution?: string;
+  }) => {
     try {
       // Handle array parsing
       if (!args.video_inputs) {
         throw new Error("Video inputs not provided");
       }
-      
+
       // Convert to array if passed as string
       let videoInputs: string[];
-      if (typeof args.video_inputs === 'string') {
+      if (typeof args.video_inputs === "string") {
         const strValue = args.video_inputs as string;
-        if (strValue.startsWith('[') && strValue.endsWith(']')) {
+        if (strValue.startsWith("[") && strValue.endsWith("]")) {
           try {
             videoInputs = JSON.parse(strValue);
           } catch {
@@ -338,43 +484,47 @@ export const geminiAnalyzeVideos = {
       } else {
         throw new Error("Invalid video_inputs: must be array or string");
       }
-      
+
       if (videoInputs.length === 0) {
         throw new Error("At least one video input must be provided");
       }
-      
+
       if (videoInputs.length > 10) {
-        throw new Error("Maximum 10 videos per request allowed for Gemini 2.5+ models");
+        throw new Error(
+          "Maximum 10 videos per request allowed for Gemini 2.5+ models"
+        );
       }
 
       // Prepare video parts for content
       const videoParts: any[] = [];
-      
+
       // Process each video input
       for (const videoInput of videoInputs) {
         const videoConfig = {
           fps: args.fps || (isYouTubeUrl(videoInput) ? 1 : 5), // Default 5 FPS for local, 1 FPS for YouTube
           startOffset: args.start_offset,
-          endOffset: args.end_offset
+          endOffset: args.end_offset,
         };
-        
+
         const videoPart = await processVideoInput(videoInput, videoConfig);
         videoParts.push(videoPart);
       }
 
       // Build content using createUserContent and createPartFromUri for uploaded files
       const contentParts: any[] = [args.prompt];
-      
+
       for (const videoPart of videoParts) {
         if (videoPart.uri && videoPart.mimeType) {
-          contentParts.push(createPartFromUri(videoPart.uri, videoPart.mimeType));
+          contentParts.push(
+            createPartFromUri(videoPart.uri, videoPart.mimeType)
+          );
         }
       }
-      
+
       const finalContents = createUserContent(contentParts);
-      
+
       const response = await ai.models.generateContent({
-        model: 'gemini-2.5-pro',
+        model: "gemini-2.5-pro",
         contents: finalContents,
       });
 
@@ -386,7 +536,7 @@ export const geminiAnalyzeVideos = {
           }
         }
       }
-      
+
       return result || "Video analysis completed but no text response received";
     } catch (error: any) {
       throw new Error(`Video analysis failed: ${error.message}`);

package/src/tools/image-to-video.ts

@@ -1,263 +1,327 @@
-import { z } from "zod";
 import * as fs from "fs";
+import { GoogleAuth } from "google-auth-library";
+import { exec } from "child_process";
 import * as path from "path";
+import { z } from "zod";
+import { getStorage } from "../storage";
 
 async function wait(ms: number): Promise<void> {
   return new Promise((resolve) => setTimeout(resolve, ms));
 }
 
-function fileToBase64(filePath: string): { data: string; mimeType: string } {
-  if (!fs.existsSync(filePath)) {
+async function fetchAccessToken(): Promise<string> {
+  try {
+    const auth = new GoogleAuth({
+      scopes: ["https://www.googleapis.com/auth/cloud-platform"],
+    });
+    const client = await auth.getClient();
+    const token = await client.getAccessToken();
+    if (!token || typeof token !== "string") {
+      throw new Error("No token from GoogleAuth");
+    }
+    return token;
+  } catch (e) {
+    // Fallback to gcloud
+    return await new Promise((resolve, reject) => {
+      exec("gcloud auth print-access-token", (err, stdout, stderr) => {
+        if (err) {
+          reject(
+            new Error(
+              `Failed to fetch an access token (ADC and gcloud): ${
+                stderr || err.message
+              }`
+            )
+          );
+          return;
+        }
+        const t = (stdout || "").trim();
+        if (!t) {
+          reject(
+            new Error(
+              "Failed to fetch an access token: empty token from gcloud"
+            )
+          );
+          return;
+        }
+        resolve(t);
+      });
+    });
+  }
+}
+
+async function fileToBase64(
+  filePath: string
+): Promise<{ data: string; mimeType: string }> {
+  const storage = getStorage();
+  if (!(await storage.exists(filePath))) {
     throw new Error(`File not found: ${filePath}`);
   }
-  const buf = fs.readFileSync(filePath);
+  const buf = await storage.readFile(filePath);
   const data = Buffer.from(buf).toString("base64");
-  // Detect mime type from extension
-  const ext = path.extname(filePath).toLowerCase();
-  const mimeType = ext === '.jpg' || ext === '.jpeg' ? 'image/jpeg' : 
-                   ext === '.png' ? 'image/png' :
-                   ext === '.webp' ? 'image/webp' : 'image/png';
+  // Default to PNG if not sure, similar to existing code
+  const mimeType = "image/png";
   return { data, mimeType };
 }
 
 export const imageToVideo = {
-  name: "imageToVideo",
-  description: "Generate videos from an image as starting first frame using Gemini Veo models via HTTP API with Gemini API key.",
+  name: "image_to_video",
+  description:
+    "Generate videos from an image as starting first frame using Vertex Veo models (predictLongRunning + fetchPredictOperation).",
   parameters: z.object({
     prompt: z.string().describe("Text description for the video"),
-    image_path: z.string().optional().describe("Path to source image for image-to-video generation"),
-    last_frame_path: z.string().optional().describe("Path to last frame image to guide ending frame (optional)"),
-    aspect_ratio: z.string().optional().describe("Video aspect ratio: '16:9' or '9:16' (default: '9:16')"),
-    duration_seconds: z.union([z.string(), z.number()]).optional().describe("Video duration in seconds: 4, 6, or 8 (default: 6)"),
-    resolution: z.string().optional().describe("Video resolution: '720p' or '1080p' (default: '720p')"),
-    negative_prompt: z.string().optional().describe("Text describing what not to include in the video"),
-    person_generation: z.string().optional().describe("Controls generation of people: 'allow_adult' (default for image-to-video) or 'allow_all'"),
-    reference_images: z.array(z.string()).optional().describe("Additional image paths for reference (max 3)"),
-    output_path: z.string().optional().describe("Output MP4 file path (if multiple predictions, index suffix is added)"),
-    gemini_api_key: z.string().optional().describe("Gemini API key (uses GEMINI_API_KEY env var if not provided)"),
-    model_id: z.string().optional().describe("Model ID (default: veo-3.1-generate-preview)"),
+    image_path: z
+      .string()
+      .optional()
+      .describe("Path to source image for image-to-video generation"),
+    last_frame_path: z
+      .string()
+      .optional()
+      .describe("Path to last frame image to guide ending frame (optional)"),
+    aspect_ratio: z
+      .string()
+      .optional()
+      .describe("Video aspect ratio: '16:9' or '9:16' (default: '9:16')"),
+    duration_seconds: z
+      .string()
+      .optional()
+      .describe("Video duration in seconds: '4', '6', or '8' (default: '6')"),
+    resolution: z
+      .string()
+      .optional()
+      .describe("Video resolution: '720p' or '1080p' (default: '720p')"),
+    negative_prompt: z
+      .string()
+      .optional()
+      .describe("Text describing what not to include in the video"),
+    person_generation: z
+      .string()
+      .optional()
+      .describe(
+        "Controls generation of people: 'allow_adult' (default for image-to-video) or 'allow_all'"
+      ),
+    reference_images: z
+      .array(z.string())
+      .optional()
+      .describe("Additional image paths for reference (max 3)"),
+    output_path: z
+      .string()
+      .optional()
+      .describe(
+        "Output MP4 file path (if multiple predictions, index suffix is added)"
+      ),
+    project_id: z
+      .string()
+      .optional()
+      .describe("GCP Project ID (default: mixio-pro)"),
+    location_id: z
+      .string()
+      .optional()
+      .describe("Vertex region (default: us-central1)"),
+    model_id: z
+      .string()
+      .optional()
+      .describe("Model ID (default: veo-3.1-fast-generate-preview)"),
+    generate_audio: z
+      .boolean()
+      .optional()
+      .describe(
+        "Boolean flag to enable generation of audio along with the video"
+      )
+      .default(false),
   }),
-  execute: async (args: {
+  async execute(args: {
     prompt: string;
     image_path?: string;
     last_frame_path?: string;
     aspect_ratio?: string;
-    duration_seconds?: string | number;
+    duration_seconds?: string;
     resolution?: string;
     negative_prompt?: string;
     person_generation?: string;
-    reference_images?: string[];
+    reference_images?: string[] | string;
     output_path?: string;
-    gemini_api_key?: string;
+    project_id?: string;
+    location_id?: string;
     model_id?: string;
-  }) => {
-    const apiKey = args.gemini_api_key || process.env.GEMINI_API_KEY;
-    if (!apiKey) {
-      throw new Error("Gemini API key is required. Set GEMINI_API_KEY environment variable or pass gemini_api_key parameter. Get one at https://aistudio.google.com/app/apikey");
-    }
+    generate_audio?: boolean;
+  }) {
+    const projectId = args.project_id || "mixio-pro";
+    const location = args.location_id || "us-central1";
+    const modelId = args.model_id || "veo-3.1-fast-generate-preview";
 
-    const modelId = args.model_id || "veo-3.1-generate-preview";
-    const baseUrl = "https://generativelanguage.googleapis.com/v1beta";
+    const token = await fetchAccessToken();
 
-    // Convert duration_seconds to number, handling both string and number inputs
-    const durationSeconds = args.duration_seconds 
-      ? (typeof args.duration_seconds === 'string' ? parseInt(args.duration_seconds) : args.duration_seconds)
-      : 6; // default
-
-    try {
-      // Build the request body for predictLongRunning
-      const instances: any[] = [
-        {
-          prompt: args.prompt,
-        },
-      ];
+    const url = `https://${location}-aiplatform.googleapis.com/v1/projects/${projectId}/locations/${location}/publishers/google/models/${modelId}:predictLongRunning`;
 
-      // Add image if provided (first frame)
-      if (args.image_path) {
-        const { data, mimeType } = fileToBase64(args.image_path);
-        instances[0].image = {
+    let imagePart: any = undefined;
+    if (args.image_path) {
+      const { data, mimeType } = await fileToBase64(args.image_path);
+      imagePart = {
+        image: {
           bytesBase64Encoded: data,
           mimeType,
-        };
-      }
+        },
+      };
+    }
 
-      // Add last frame if provided (for interpolation)
-      if (args.last_frame_path) {
-        const { data, mimeType } = fileToBase64(args.last_frame_path);
-        instances[0].lastFrame = {
+    let lastFramePart: any = undefined;
+    if (args.last_frame_path) {
+      const { data, mimeType } = await fileToBase64(args.last_frame_path);
+      lastFramePart = {
+        lastFrame: {
           bytesBase64Encoded: data,
           mimeType,
-        };
-      }
-
-      // Add reference images if provided
-      if (args.reference_images && args.reference_images.length > 0) {
-        const refImages = args.reference_images.slice(0, 3).map((imgPath) => {
-          const { data, mimeType } = fileToBase64(imgPath);
-          return {
-            image: {
-              bytesBase64Encoded: data,
-              mimeType,
-            },
-            referenceType: "asset",
-          };
-        });
-        instances[0].referenceImages = refImages;
-      }
-
-      // Build parameters - NOTE: Parameters go in "parameters" object, not in instances
-      const parameters: any = {};
-
-      if (args.aspect_ratio) {
-        parameters.aspectRatio = args.aspect_ratio;
-      }
-      
-      if (durationSeconds) {
-        parameters.durationSeconds = durationSeconds;
-      }
-      
-      if (args.resolution) {
-        parameters.resolution = args.resolution;
-      }
+        },
+      };
+    }
 
-      if (args.negative_prompt) {
-        parameters.negativePrompt = args.negative_prompt;
+    let referenceImages: any[] | undefined = undefined;
+    if (args.reference_images) {
+      let refImages: string[];
+      if (typeof args.reference_images === "string") {
+        if (
+          args.reference_images.startsWith("[") &&
+          args.reference_images.endsWith("]")
+        ) {
+          try {
+            refImages = JSON.parse(args.reference_images);
+          } catch {
+            throw new Error("Invalid reference_images format");
+          }
+        } else {
+          refImages = [args.reference_images];
+        }
+      } else if (Array.isArray(args.reference_images)) {
+        refImages = args.reference_images;
+      } else {
+        throw new Error("Invalid reference_images: must be array or string");
       }
 
-      if (args.person_generation) {
-        parameters.personGeneration = args.person_generation;
+      if (refImages.length > 0) {
+        referenceImages = await Promise.all(
+          refImages.slice(0, 3).map(async (p) => {
+            const { data, mimeType } = await fileToBase64(p);
+            return {
+              image: {
+                bytesBase64Encoded: data,
+                mimeType,
+              },
+              referenceType: "asset",
+            };
+          })
+        );
       }
+    }
 
-      const requestBody: any = {
-        instances,
-      };
+    const personGeneration =
+      args.person_generation || (args.image_path ? "allow_adult" : "allow_all");
+
+    const instances: any[] = [
+      {
+        prompt: args.prompt,
+        ...(imagePart || {}),
+        ...(lastFramePart || {}),
+        ...(referenceImages ? { referenceImages } : {}),
+      },
+    ];
+
+    const parameters: any = {
+      aspectRatio: args.aspect_ratio || "9:16",
+      durationSeconds: parseInt(args.duration_seconds || "6") || 6,
+      resolution: args.resolution || "720p",
+      negativePrompt: args.negative_prompt,
+      generateAudio: args.generate_audio || false,
+      personGeneration,
+    };
+
+    const res = await fetch(url, {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${token}`,
+        "Content-Type": "application/json",
+      },
+      body: JSON.stringify({ instances, parameters }),
+    });
+
+    if (!res.ok) {
+      const text = await res.text();
+      throw new Error(`Vertex request failed: ${res.status} ${text}`);
+    }
 
-      // Only add parameters if we have any
-      if (Object.keys(parameters).length > 0) {
-        requestBody.parameters = parameters;
-      }
+    const op = (await res.json()) as any;
+    const name: string = op.name || op.operation || "";
+    if (!name) {
+      throw new Error(
+        "Vertex did not return an operation name for long-running request"
+      );
+    }
 
-      console.log(`Starting video generation with model: ${modelId}`);
+    let current = op;
+    let done = !!op.done;
+    let tries = 0;
 
-      // Step 1: Start the long-running operation
-      const url = `${baseUrl}/models/${modelId}:predictLongRunning`;
-      const response = await fetch(url, {
+    // Poll using fetchPredictOperation as per Vertex recommendation
+    const fetchUrl = `https://${location}-aiplatform.googleapis.com/v1/projects/${projectId}/locations/${location}/publishers/google/models/${modelId}:fetchPredictOperation`;
+    while (!done && tries < 60) {
+      await wait(10000);
+      const poll = await fetch(fetchUrl, {
         method: "POST",
         headers: {
-          "x-goog-api-key": apiKey,
+          Authorization: `Bearer ${token}`,
           "Content-Type": "application/json",
         },
-        body: JSON.stringify(requestBody),
+        body: JSON.stringify({ operationName: name }),
       });
-
-      if (!response.ok) {
-        const errorText = await response.text();
-        throw new Error(`Video generation request failed: ${response.status} ${errorText}`);
-      }
-
-      const operation = await response.json() as any;
-      const operationName: string = operation.name || operation.operation || "";
-      
-      if (!operationName) {
-        throw new Error("No operation name returned from API");
-      }
-
-      console.log(`Operation started: ${operationName}`);
-
-      // Step 2: Poll the operation status by getting the operation directly
-      let currentOp: any = operation;
-      let done = !!operation.done;
-      let tries = 0;
-      const maxTries = 60; // 10 minutes with 10s intervals
-
-      while (!done && tries < maxTries) {
-        await wait(10000); // Wait 10 seconds
-        tries++;
-        console.log(`Polling attempt ${tries}/${maxTries}...`);
-
-        // Poll by getting the operation status directly
-        const pollResponse = await fetch(`${baseUrl}/${operationName}`, {
-          method: "GET",
-          headers: {
-            "x-goog-api-key": apiKey,
-          },
-        });
-
-        if (!pollResponse.ok) {
-          const errorText = await pollResponse.text();
-          throw new Error(`Operation polling failed: ${pollResponse.status} ${errorText}`);
-        }
-
-        currentOp = await pollResponse.json() as any;
-        done = !!currentOp.done || !!currentOp.response;
-      }
-
-      if (!done) {
-        throw new Error("Video generation timed out after 10 minutes");
+      if (!poll.ok) {
+        const text = await poll.text();
+        throw new Error(`Vertex operation poll failed: ${poll.status} ${text}`);
       }
+      current = (await poll.json()) as any;
+      done = !!current.done || !!current.response;
+      tries++;
+    }
 
-      console.log("Operation completed!");
-
-      // Step 3: Extract and download videos
-      const resp = currentOp.response || currentOp;
-      
-      // The response structure is: response.generateVideoResponse.generatedSamples[].video.uri
-      const generateVideoResponse = resp?.generateVideoResponse;
-      const generatedSamples = generateVideoResponse?.generatedSamples || [];
-      
-      if (!generatedSamples || generatedSamples.length === 0) {
-        let jsonStr = "";
-        try { jsonStr = JSON.stringify(resp, null, 2); } catch {}
-        return `Video generation completed but no generatedSamples found.\n\nFull Response:\n${jsonStr.slice(0, 1000)}${jsonStr.length > 1000 ? '\n...(truncated)' : ''}`;
-      }
-
-      const outputs: string[] = [];
-
-      // Download videos from URIs
-      for (let i = 0; i < generatedSamples.length; i++) {
-        const sample = generatedSamples[i];
-        const videoUri = sample?.video?.uri;
-        
-        if (!videoUri) {
-          console.warn(`Sample ${i} has no video URI`);
-          continue;
-        }
-
-        console.log(`Downloading video ${i + 1}/${generatedSamples.length} from ${videoUri}...`);
-        
-        // Download video from URI with API key
-        const videoResponse = await fetch(videoUri, {
-          method: "GET",
-          headers: {
-            "x-goog-api-key": apiKey,
-          },
-        });
-
-        if (!videoResponse.ok) {
-          throw new Error(`Failed to download video: ${videoResponse.status} ${videoResponse.statusText}`);
+    const resp = current.response || current;
+    // Decode from response.videos[].bytesBase64Encoded only
+    const outputs: string[] = [];
+    const saveVideo = async (base64: string, index: number) => {
+      if (!base64) return;
+      const filePath = args.output_path
+        ? index === 0
+          ? args.output_path
+          : args.output_path.replace(/\.mp4$/i, `_${index}.mp4`)
+        : `video_output_${Date.now()}${index === 0 ? "" : "_" + index}.mp4`;
+      // For storage provider, we use the path as is (relative or absolute)
+      // If using LocalStorage, it handles resolving.
+      // If using Payload, it handles the key.
+
+      const buf = Buffer.from(base64, "base64");
+      const storage = getStorage();
+      await storage.writeFile(filePath, buf);
+      outputs.push(filePath);
+    };
+
+    if (Array.isArray(resp?.videos) && resp.videos.length > 0) {
+      for (let i = 0; i < resp.videos.length; i++) {
+        const v = resp.videos[i] || {};
+        if (typeof v.bytesBase64Encoded === "string") {
+          await saveVideo(v.bytesBase64Encoded, i);
         }
-
-        const videoBuffer = await videoResponse.arrayBuffer();
-        
-        // Save video to file
-        const filePath = args.output_path
-          ? (i === 0 ? args.output_path : args.output_path.replace(/\.mp4$/i, `_${i}.mp4`))
-          : `video_output_${Date.now()}${i === 0 ? '' : '_' + i}.mp4`;
-        const absPath = path.resolve(filePath);
-        
-        fs.writeFileSync(absPath, Buffer.from(videoBuffer));
-        outputs.push(absPath);
-        console.log(`Saved video to: ${absPath}`);
-      }
-
-      if (outputs.length > 0) {
-        return `Video(s) saved successfully:\n${outputs.map((p, i) => `${i + 1}. ${p}`).join('\n')}`;
       }
-
-      return "Video generation completed but no videos were saved.";
-      
-    } catch (error: any) {
-      throw new Error(`Video generation failed: ${error.message || JSON.stringify(error)}`);
     }
+    if (outputs.length > 0) {
+      return `Video(s) saved: ${outputs.join(", ")}`;
+    }
+
+    // If nothing saved, return a concise summary plus head/tail snippets of JSON
+    let jsonStr = "";
+    try {
+      jsonStr = JSON.stringify(resp);
+    } catch {}
+    const head150 = jsonStr ? jsonStr.slice(0, 150) : "";
+    const tail50 = jsonStr
+      ? jsonStr.slice(Math.max(0, jsonStr.length - 50))
+      : "";
+    return `Vertex operation done but no videos array present. operationName=${name}. json_head150=${head150} json_tail50=${tail50}`;
   },
 };