Last September, Release Notes 231 announced the Icecat MCP Server in Beta. At the time, it could do exactly one thing: return product information when an AI agent supplied a product code and a brand name.
It now offers nineteen tools; it is listed in the official public MCP Registry; and, more interestingly, we finally have enough real traffic to answer the question we are asked most often: what do people actually do with it?
We went through our usage statistics account by account and, over many months, spoke with the partners behind them. This article shares what we found: five use cases, several of which we did not anticipate, and an honest look at the numbers after almost a year.
Monthly requests to the MCP server, September 2025 through 15 August 2026.
Five honest observations about this chart.
Growth comes in spikes, not a smooth curve. Each of the large jumps corresponds to a partner switching on an integration, not to a gradual rise in demand. July 2026 was our largest month, with just over 17,000 requests (five times the previous record), and roughly 59% of those requests came from a single account. Some of those spikes then decay as a project finishes.
The number of active companies grew steadily: from four in September 2025 to between 20 and 30 every month through the summer of 2026. We pay closer attention to this figure than to the total number of requests because it shows how many partners are using the server, rather than how much traffic a single large integration generates.
Weekly engagement has roughly tripled. Counting companies that make meaningful use of the server in a given week, we have moved from about four or five per week in spring to 10–15 per week across July and August.
Early months were deliberately constrained. In Release 232, we temporarily limited MCP access to Open Icecat data while we studied the traffic. Low autumn volumes reflect a narrowed product, not weak interest.
New integrations can scale quickly. August 2026 in the chart covers only the first fifteen days, and is running at a similar daily rate to July — but driven by an entirely different partner, one that went from around 70 calls on a single day to over 1,600 two days later. Once a workflow works, volume arrives fast.
One of our highest-volume partners is a retailer whose local language is not currently one of Icecat’s locales. They pull the full product overview through MCP and translate it themselves as part of an automated workflow.
This was not a use case we designed for, and it demonstrates something useful: the MCP server works well as a factual grounding source for translation pipelines, because everything the model translates comes from structured product data rather than from scraped marketing copy.
It is worth adding, for anyone considering the same approach, that Icecat already offers a dedicated localization solution for partners who need high-quality translated product content. That approach is already adopted by multiple partners and does not require you to build a translation pipeline yourself. If that is of interest to your catalog, speak to your Icecat account manager or get in touch via our contact form.
Several partners use the MCP server for something deliberately unglamorous: retrieving product content for their own catalog, exactly as they would through the Icecat API.
This is worth stating plainly, because it is probably the most widely applicable use case of all. MCP is an alternative interface to the same product data. If your team already works with an AI assistant, you do not need to write and maintain an API client to get product content into a workflow. You point the assistant at the server and ask it for what you need. Specifications, images, descriptions, and identifiers all come back as structured data.
For teams without dedicated development capacity, this considerably lowers the barrier to using Icecat content.
An advertising agency used the server in two concentrated bursts and then stopped. This shape we now recognize as a client project rather than an ongoing process. The use case: take real product specifications and generate optimized, unique content around them, linked to paid search campaigns.
A second retailer uses MCP to generate review-style content – the genuine pluses and minuses of a product, derived from its actual specifications – in order to publish unique material and attract organic search traffic.
Both are the same insight from different angles. Specifications are facts, and facts are what generative models need in order to write something true. MCP is a convenient way to hand a model the facts.
Many of our partners sell through marketplaces, and every marketplace has its own publication standards. Amazon is the obvious example, but the same applies to Shopee, Lazada and others. Title lengths, bullet formats, mandatory attributes, and prohibited wording all differ from platform to platform, and from the standardized product content a partner holds in their own systems.
Reshaping content to fit each platform used to be manual work. AI handles reshaping well, but only if it works from facts. Ask a model to rewrite an existing marketplace listing, and it will confidently invent specifications that were never true, because its starting point was somebody else’s marketing copy rather than the product.
This is where MCP fits. It supplies the verified facts: real specifications, real identifiers, real images. An AI workflow can transform standardized product content into each marketplace’s required format without inventing anything along the way. The sequence is: retrieve the facts, then reshape to the platform’s rules.
The principle generalizes well beyond marketplaces: generative content pipelines need a factual anchor, and product data is that anchor. Without one, content generation slides into content invention.
Lena, Icecat’s AI Assistant, launched on Icecat.biz in Release 254 and is herself an MCP consumer. She queries the server to answer customer questions about products.
We built Lena for our own support operation rather than as a showcase, but she has become one of the most instructive accounts on the server. She calls twelve different tools, more than almost any partner, and her usage mix looks unlike most: she leads with image galleries and product identifiers rather than the general product overview, because that is what a support conversation actually needs.
The most striking thing is consistency. Lena has been active on nearly every day since launch, which is what makes an agent integration different in kind from a human workflow: once deployed, it simply runs. It does not forget the integration exists or get pulled onto another project.
Product search. Every tool we offer today assumes you already know which product you want. Search is what lets an agent find a product it cannot yet name, and it is the largest remaining gap. It is currently in development.
Multilingual output through MCP. Our partners are multilingual, and we are actively improving how the server handles locales in its responses.
Simpler connection. Connecting an AI client to the MCP server currently requires editing a configuration file by hand, whereas many connectors can be installed in a couple of clicks via a standard authorization flow. Making that first connection easier is high on our list because the quicker a workflow starts, the sooner it becomes part of someone’s routine.
A year ago, the Icecat MCP Server could answer one question. It now supports translation workflows, straightforward content retrieval, SEO content generation, marketplace publishing and AI support assistants.
The most consistent consumer of Icecat product data on the server is not a person; it is an AI. If that generalizes, and we increasingly think it will, a meaningful share of product content consumption is shifting from humans browsing to agents querying. Agents do not care about a pleasant interface. They care enormously about product content that is complete, well-structured, and predictable, which is, conveniently, what we have spent twenty years building.
If you use the Icecat MCP Server, we would like to hear what for. The use cases above took months of conversations with partners to piece together, and each one taught us something that statistics alone could never have shown us. Get in touch with your account manager, or reply to this post.
Read further: Icecat, e-commerce, ecommerce, Icecat, MCP, product content
