Add MCP server and enhance AI content tools

Implement self-hosted MCP server for admin actions, update AI generation tools in admin dashboard, and refine mobile app stores and components.

apps/admin/src/lib/ai/analyzers/props.ts

@@ -0,0 +1,89 @@
+// =============================================================================
+// Props Analysis Functions
+// Identify required props/accessories for questions
+// =============================================================================
+
+import { getOpenAI, getModel, getTemperature } from '../client';
+import type { PropsAnalysis } from '../types';
+
+/**
+ * Analyze questions for required props/accessories
+ */
+export async function analyzeQuestionProps(
+    questions: {
+        id: string;
+        text: string;
+        partner_text?: string | null;
+        required_props?: string[] | null;
+    }[],
+    existingProps: string[] = []
+): Promise<PropsAnalysis[]> {
+    const openai = getOpenAI();
+
+    const normalizedExistingProps = existingProps
+        .map(prop => prop.trim())
+        .filter(Boolean);
+
+    const simplifiedQuestions = questions.map(q => ({
+        id: q.id,
+        text: q.text,
+        partner_text: q.partner_text,
+        current_required_props: q.required_props ?? null,
+    }));
+
+    const prompt = `<task>
+Identify any props/accessories required to perform each question.
+Only include questions where required props should change (missing, incorrect, or overly generic).
+</task>
+
+<existing_props>
+Prefer these exact prop names when possible (reuse existing names instead of inventing new ones):
+${JSON.stringify(normalizedExistingProps)}
+</existing_props>
+
+<rules>
+1. Props are physical items/accessories, not body parts or acts.
+2. Use lowercase, short, singular names (e.g., "blindfold", "remote vibrator").
+3. If a prop is essentially the same as an existing one, reuse the existing name.
+4. Avoid making up new variants when a close existing name is available.
+5. Example: if "remote vibrator" exists, map "wifi" or "app-controlled" variants to "remote vibrator".
+6. Use specific items when clearly stated; otherwise allow "toy" or "props".
+7. If no props are required, set suggested_required_props to null.
+8. Only return entries where the suggested props differ from current_required_props.
+</rules>
+
+<questions_to_analyze>
+${JSON.stringify(simplifiedQuestions)}
+</questions_to_analyze>
+
+<output_format>
+{
+  "analysis": [
+    {
+      "id": string,
+      "suggested_required_props": string[]|null,
+      "reason": string
+    }
+  ]
+}
+</output_format>`;
+
+    const response = await openai.chat.completions.create({
+        model: getModel('fix'),
+        messages: [
+            {
+                role: 'system',
+                content: 'You are a content analyst for a couples app. Identify required props/accessories and return valid JSON.'
+            },
+            { role: 'user', content: prompt }
+        ],
+        response_format: { type: 'json_object' },
+        temperature: getTemperature('fix', 0.3),
+    });
+
+    const content = response.choices[0].message.content;
+    if (!content) throw new Error('No content generated');
+
+    const parsed = JSON.parse(content);
+    return parsed.analysis || [];
+}