Enable Product Discovery in ChatGPT

Dynamic Yield enables you to connect Shopping Muse to ChatGPT through a managed MCP integration. The integration is exposed in ChatGPT as a plugin and provides conversational product discovery directly within the chat experience.

Dynamic Yield generates the MCP server URL in Experience OS, and uses your existing Dynamic Yield product feed, recommendation logic, and Shopping Muse configuration, so you don’t need to build or host your own MCP infrastructure. After connecting the MCP server to your ChatGPT integration, ChatGPT can send shopper queries to Dynamic Yield, retrieve structured results, and present them directly in the conversation.

The core components of the integration are:

• ChatGPT plugin: The shopper-facing integration that users discover and install from ChatGPT.
• MCP server (Dynamic Yield): The managed integration layer connecting ChatGPT to Dynamic Yield services.
• Shopping Muse: The conversational product discovery engine.
• ExperienceOS ChatGPT app: The Dynamic Yield app used to set up the integration and view reports.

The integration flow

When the integration is operational, this is how it works:

1. In ChatGPT:
   1. A shopper asks a product-related question using your ChatGPT plugin.
   2. ChatGPT calls the Dynamic Yield search tool via MCP.
2. The search process:
   1. Dynamic Yield sends the request to Shopping Muse.
   2. Shopping Muse returns relevant product recommendations.
   3. The MCP server formats the response for ChatGPT.
   4. ChatGPT displays the returned products in structured widgets in the conversation.
3. A dedicated tool automatically tracks impressions and clicks.

MCP tools

MCP tools are structured capabilities that ChatGPT invokes through the MCP server during a conversation. They enable ChatGPT to retrieve product recommendations and track shopper engagement using Dynamic Yield services.

These tools are fully managed by Dynamic Yield, with no additional development work required. The following descriptions and schema are provided for reference.

Search

Used for product discovery when the ChatGPT user shows shopping intent.

Query examples:

• "Black dresses under $200"
• "Running shoes for flat feet"

Input:

1     { 
2       "query": "women black evening dress under 300 ILS size M" 
3     } 

Output (simplified):

1     { 
2       "assistant": "Here are a few options", 
3       "widgets": [ 
4         { 
5           "title": "Black Dresses", 
6           "slots": [ 
7             { 
8               "slotId": "slot-123", 
9               "productData": { 
10                 "name": "Classic Black Dress", 
11                 "price": 149.99, 
12                 "image_url": "https://example.com/image.jpg" 
13               } 
14             } 
15           ] 
16         } 
17       ] 
18     } 

Engagement

Automatically tracks shopper interactions with recommended products:

• Product impressions
• Product clicks

1     { 
2       "slotId": "slot-123" 
3     }

UI rendering

Product results are rendered directly inside ChatGPT as interactive widgets, including:

• Product cards (image, name, price)
• Grouped results (widgets)
• Supporting text from the assistant

Dynamic Yield controls the widget UI, which runs in a sandboxed environment.

Security and privacy

• Each integration uses a unique MCP Server URL
• API keys are handled server-side only
• Requests are routed securely through Dynamic Yield infrastructure
• User data is isolated at the section level