Can You Use products.ai in ChatGPT or Claude? Connecting Your AI With MCP

What is MCP, in plain words?

MCP, short for Model Context Protocol, is an open standard for connecting AI assistants to outside tools and data. Anthropic introduced it in November 2024, and the specification is public at modelcontextprotocol.io.

Before a shared standard, every pairing of an AI app and a tool needed its own custom integration. MCP splits the job in two. A provider runs an MCP server that describes its tools. An AI app that speaks the protocol, called an MCP client, connects to that server and calls those tools when a question needs them.

For shopping, that split makes sense. Your AI is good at understanding what you mean. A product source knows which products exist, what owners say about them and which shops have them in stock.

Why connect products.ai instead of letting your AI search the web?

Because a general chatbot can mix up similar models or repeat details from an old page. products.ai matches the exact product on its barcode (GTIN), weighs it against what you asked, and names a shop that has it in stock, with the date that shop listed it.

Model mix-ups are the sneaky part. The Sony WH-1000XM4 and WH-1000XM5 differ by one digit, yet they are different headphones. A page that just says “Sony noise-cancelling headphones” can send your AI to the wrong one. A barcode settles it, because each GTIN belongs to one specific product.

Web search still has its place, and products.ai reads reviews and forums too. For where to buy, a dated shop listing beats a page of unknown age.

What does your AI get from products.ai through MCP?

It gets what Pius Bot would answer, as data: one pick that fits what you asked, why it fits, a one-line why not for each alternative, and where to buy it.

Ask for noise-cancelling headphones under €300 for travel and the Sony WH-1000XM5 comes back as the best fit for flights. The Sennheiser Momentum 4 gets its why not (longest battery, but less quiet on a plane), and so does the Bose QC Ultra (€35,85 over budget). As the place to buy, products.ai names bol.com: €226,99, in stock, delivered in 1–2 days.

Shops in your country come first, drawn from 448 merchants across 20 countries and 11 currencies.

How do you connect Claude or ChatGPT to products.ai?

Request access, and your MCP URL comes with it. Access is by request for now, a person in Amsterdam reads every request, and the connection works in any MCP client: Claude, ChatGPT, Gemini and others.

The request form asks for a work email, your company or project, and what you’re interested in. The box for what you’re building is optional, but fill it in. A line like “travel gear picks inside our trip planner, for the UK and Germany” tells the reader far more than an empty field. The button opens a pre-filled email, and nothing is sent until you press send. Or write to [email protected].

Setup differs per platform, so we don’t publish click-by-click steps. Menus move between versions. Broadly, you paste the URL into an app that accepts one, install the extension or connector where your platform offers one, or add it to a config file in developer tools. Connect your AI shows that file’s shape. If your app isn’t named but speaks MCP, say so in your request.

What is the difference between In your AI, In your product and Pius Bot?

They are three ways into the same data: chat for shoppers who ask Pius Bot directly, In your AI for the assistant you already use, and In your product for builders who put products.ai inside their own app.

Our take: if you only want answers, start with chat, because it needs the least setup. If you already spend your day in Claude or ChatGPT, connect it there. If you ship software to other people, In your product is built for that. The plans sit side by side in pricing.

What data do builders get from the API or MCP tools?

Picks with reasons, as structured data, with every product matched by barcode. The tool names and fields below are illustrative until the docs ship, so treat them as a preview.

• search_products: products for a plain-language question.
• get_product: one barcode, with specs, variants and alternatives.
• compare_products: products side by side, by barcode.
• get_offers: listed offers with price, stock and date.

A response carries best_fit, why_it_fits, a why_not per alternative, a best_pick (the shop to buy from, with its price and the reason) and a second_chance list. In the headphones example, best_pick is bol.com at €226,99, in stock and delivered in 1–2 days. A refurbished grade A pair at €179,00 lands in second_chance, because the question asked for new.

Keep the reasons visible in your app. They show your users why this product and this shop, and that is what makes the next pick believable. The API overview shows a full sample.

How does commission work, and can your app earn from it?

Yes, where a merchant runs an affiliate program. The link is flagged in the data, you disclose it in your app, and a sale through it can earn your app a share of the commission. Terms vary per program.

Commission never picks the product. Pius Bot picks on fit first, before it looks at any shop, and affiliate links are added after the shop is chosen. Shops can’t pay to be recommended. A shop gets named by having the exact product in stock, delivering fast and making returns easy.

That order protects you as a builder. If the pick followed the money, your users would notice, and a recommendation nobody trusts earns nothing. The full reasoning is in how Pius Bot chooses a product.