Product Discovery & Catalogue Agent

Expose your catalogue as MCP search tools

Create MCP tools that give the agent flexible search and retrieval access to your product catalogue.

Tools to expose via MCP:
- search_products(query, filters?) → returns matching products
  filters: category, price_min, price_max, weight, colour, in_stock
- get_product_details(product_id) → full spec sheet, images, reviews
- get_related_products(product_id) → frequently bought together
- check_stock(product_id, location?) → real-time availability
- get_categories() → full category tree for context

Include rich product metadata

The richer your product data, the better the agent performs. Include: dimensions, weight, use-case tags, material, target audience, key features. The agent uses all of this to match intent — not just product names.

Register in SentientOne

Add the catalogue MCP server in the SentientOne dashboard. You can restrict this agent to read-only catalogue tools — no account or order data needed for discovery.

Create the Product Discovery Agent

Write a system prompt that turns the agent into a knowledgeable sales assistant for your domain.

System Prompt:
You are an expert product advisor for [Store Name], a [category] retailer.
Your goal is to help customers find exactly the right product for their needs.

When a customer describes what they want:
1. Use search_products to find matching items based on their requirements
2. If multiple options exist, present the top 2-3 with key differentiators
3. Highlight specs that are directly relevant to their stated need
4. If stock is low, mention it to create appropriate urgency
5. Always offer to narrow down further: "Would you like options under $X?" or "Do you need it to be [feature]?"

Tone: Knowledgeable, friendly, not pushy. Like the best sales assistant in the store.

Attach catalogue MCP tools

Connect the product catalogue MCP server. The agent will autonomously call search_products, get_product_details, and check_stock as needed to build a complete answer.

Optional: add purchase history context

For logged-in users, also connect the user profile tool. The agent can then avoid recommending products they already own and factor in their stated preferences from past purchases.

Add an AI search bar or chat button

Replace or supplement your keyword search with an AI query input:

// Search bar → Agent API
async function aiSearch(query: string, userId?: string) {
  const res = await fetch("https://app.sentientone.ai/api/chat", {
    method: "POST",
    headers: {
      "X-Agent-Id": "YOUR_PRODUCT_DISCOVERY_AGENT_ID",
      "X-App-Key": "YOUR_APP_KEY"
    },
    body: JSON.stringify({
      message: query,
      sessionId: userId || `anon-${Date.now()}`
    })
  });
  const { reply } = await res.json();
  return reply; // render in your search results area
}

Add a floating chat assistant

Add a floating 'Ask our AI' button on product listing pages. The agent can help users navigate from broad searches to specific products through conversation — matching the experience of a knowledgeable sales floor assistant.

Track and improve

Use SentientOne analytics to see which queries aren't finding good matches — these are signals to improve your product data, add new categories, or expose additional MCP tools. Every conversation is a source of product discovery insight.