fix svelte warning & fix youtube and uploaded images for MCP

Author: 0xGingi

src/lib/backend/mcp-tools.ts

@@ -206,14 +206,14 @@ export const mcpToolDefinitions: ChatCompletionTool[] = [
                     },
                     image_url: {
                         type: 'string',
-                        description: 'URL of the image to analyze'
+                        description: 'URL of the image to analyze. If omitted, the most recent image in the chat will be used.'
                     },
                     model: {
                         type: 'string',
                         description: 'Optional vision model to use (default: glm-4.6v)'
                     }
                 },
-                required: ['prompt', 'image_url']
+                required: ['prompt']
             }
         }
     }
@@ -387,7 +387,7 @@ async function executeYoutubeTranscribe(
             'Authorization': `Bearer ${apiKey}`,
             'Content-Type': 'application/json'
         },
-        body: JSON.stringify({ url })
+        body: JSON.stringify({ urls: [url] })
     });
 
     if (!response.ok) {
@@ -399,7 +399,21 @@ async function executeYoutubeTranscribe(
 
     let resultText = `YouTube transcript for: ${url}\n\n`;
 
-    if (data.transcript) {
+    // Handle new API response format (array of transcripts)
+    if (data.transcripts && Array.isArray(data.transcripts) && data.transcripts.length > 0) {
+        const result = data.transcripts[0];
+        if (result.success && result.transcript) {
+            const transcript = result.transcript.slice(0, 8000);
+            resultText += transcript;
+            if (result.transcript.length > 8000) {
+                resultText += `\n... (transcript truncated, ${result.transcript.length} chars total)`;
+            }
+        } else {
+            return { success: false, result: '', error: result.error || 'Failed to retrieve transcript' };
+        }
+    }
+    // Fallback for potential legacy/singular response format
+    else if (data.transcript) {
         const transcript = data.transcript.slice(0, 8000);
         resultText += transcript;
         if (data.transcript.length > 8000) {
@@ -411,6 +425,8 @@ async function executeYoutubeTranscribe(
         if (data.text.length > 8000) {
             resultText += `\n... (transcript truncated, ${data.text.length} chars total)`;
         }
+    } else {
+        return { success: false, result: '', error: 'No transcript found in response' };
     }
 
     return { success: true, result: resultText };

src/routes/account/+page.svelte

@@ -654,10 +654,11 @@
 
 					<div class="flex flex-col gap-2">
 						<div class="flex justify-between">
-							<label class="text-sm leading-none font-medium">Speed</label>
+							<label for="tts-speed" class="text-sm leading-none font-medium">Speed</label>
 							<span class="text-muted-foreground text-xs">{localSettings.ttsSpeed}x</span>
 						</div>
 						<input
+							id="tts-speed"
 							type="range"
 							min="0.25"
 							max="4.0"

src/routes/api/generate-message/+server.ts

@@ -908,7 +908,42 @@ ${attachedRules.map((r) => `- ${r.name}: ${r.rule}`).join('\n')}`;
 				if (!tc.id || !tc.function.name) continue;
 
 				try {
-					const args = JSON.parse(tc.function.arguments || '{}');
+					let args = JSON.parse(tc.function.arguments || '{}');
+
+					// Special handling for vision tool: inject image from context if missing
+					if (tc.function.name === 'nanogpt_vision') {
+						const imgUrl = args.image_url as string | undefined;
+						// If no URL provided, or it's not a valid URL/data-uri, try to find one in context
+						if (!imgUrl || (!imgUrl.startsWith('http') && !imgUrl.startsWith('data:'))) {
+							log('Background: Vision tool called without valid URL, searching context for images', startTime);
+
+							// Search backwards for the most recent image
+							let foundImage: string | null = null;
+							// Iterate formattedMessages in reverse
+							for (let i = formattedMessages.length - 1; i >= 0; i--) {
+								const msg = formattedMessages[i];
+								if (!msg) continue;
+								if (typeof msg.content !== 'string' && Array.isArray(msg.content)) {
+									for (const part of msg.content) {
+										if (part && typeof part === 'object' && 'type' in part && part.type === 'image_url' && 'image_url' in part) {
+											// @ts-ignore - TS might not infer the type perfectly here
+											foundImage = part.image_url.url;
+											break;
+										}
+									}
+								}
+								if (foundImage) break;
+							}
+
+							if (foundImage) {
+								log('Background: Injected image from context into vision tool', startTime);
+								args.image_url = foundImage;
+							} else {
+								log('Background: No image found in context for vision tool', startTime);
+							}
+						}
+					}
+
 					log(`Background: Executing tool ${tc.function.name}`, startTime);
 
 					const result = await executeMcpTool(tc.function.name, args, apiKey);