Do gumtxxx and okxxxx Need MCP? Start by Helping Buyers Find Things
Should classifieds platforms build an MCP server for sellers or shoppers? A practical look at AI shopping demand, listing quality and useful enquiries.
Like it ? share it
Looking for a used bicycle usually means opening a few sites, setting a budget and location, then reading through frame sizes, condition notes and seller descriptions. An AI assistant lets the buyer start with something more natural:
Find me a used commuter bike for under £200 that I can collect nearby this weekend.
The request is easy to understand. The harder questions come next. Is the bike still available? Will it fit? Can the seller arrange collection on Saturday?
For classifieds platforms such as gumtxxx and okxxxx, that is where AI shopping could be useful. The platform has the listings and seller information. The assistant helps turn a buyer's needs into a shortlist.
Whether that calls for an MCP server depends on what the connection actually helps people do.
People will ask AI for advice before they trust it to buy
AI already plays a part in shopping. A 2026 Retail Economics and Barclays survey found that 38% of UK consumers had used conversational AI for shopping tasks. Adobe's March 2026 US survey reported 39% had used AI for online shopping. The surveys use different methods, so they do not establish which country is ahead. Both suggest there is an audience worth taking seriously. UK research, US research.
Trust is more limited. In YouGov's US research, 65% trusted AI to compare prices, while only 14% trusted it to place orders. YouGov.
That gap makes sense in second-hand shopping. A dining table might be cheap, but the buyer still needs to judge the scratches, check whether the legs come off and work out how to get it home.
An assistant can make those decisions easier by narrowing the options and explaining what still needs checking with the seller. That is already a useful service.
Fast traffic growth still needs a business case
Adobe recorded a 393% year-on-year increase in AI referrals to US retail sites in the first quarter of 2026. That is a striking growth rate, but it does not tell us what share of all visits came from AI. Adobe.
For a separate view of scale, Ahrefs' public dashboard reported that ChatGPT referrals accounted for 0.32% of traffic in its August 2026 sample. This covers a broad set of websites; it is not a classifieds benchmark or an estimate for either platform. Ahrefs.
There is enough evidence to test AI as a new source of customers. There is much less reason to divert investment from search and the existing product on the assumption that AI will replace them.
Follow what buyers do after seeing a recommendation. They may click through, contact a seller and receive a reply. Each step tells you something different. Citations and tool calls alone can produce an impressive dashboard without showing whether anyone did more business.
An MCP server needs somewhere people will use it
The Model Context Protocol, or MCP, gives AI assistants a standard way to call external tools. A marketplace can expose listing search, item details and availability checks. With the right permissions, it can also let sellers update stock or edit adverts.
The connection still needs users. A partner's assistant might send buyers to the listings, or professional sellers might use it through a tool they already work in. Without a clear route to either group, publishing an endpoint is unlikely to create demand on its own.
MCP is also one of several ways to connect. OpenAI's merchant documentation starts with structured product feeds. Google's Universal Commerce Protocol supports integration through APIs and MCP, among other options. Find out what a prospective partner needs before choosing the interface. OpenAI documentation, Google documentation.
Much of the valuable work is less visible: keeping prices current, removing sold items, checking locations and making sure photos match the listing. A working interface is little comfort to someone who travels to collect an item that has already gone.
Illustration: AI helps narrow the options; the buyer checks the details with the seller.
Start with the side of the market that has a clear task
Seller tools are a sensible place to look for repetitive work. A used-furniture shop might spend time taking sold stock offline, editing new listings and sorting buyer questions. Time saved and missed updates are relatively easy to measure.
If the existing dashboard already handles those tasks well, adding an AI connection may achieve little. Find the slow or frustrating step first.
For buyers, search is a reasonable starting point. Someone asking for a dining table for a small living room can be prompted for dimensions, budget and collection distance. The platform then needs to return suitable, available items and make it easy to contact the seller.
A platform with an existing AI search integration should examine how well that journey works. A platform starting from scratch could work with one partner on one category, such as furniture or bicycles. Cars, property, jobs and local services all require different information; supporting them can come later.
Public search also needs a different permission boundary from editing adverts, accessing account information or sending messages. Offering the first does not require opening the rest.
A small trial can answer the useful questions
I would begin with one category and a narrow set of functions: search, listing details and a handoff to the normal contact process.
For buyers, track whether the results fit their needs, whether they make worthwhile enquiries and whether sellers reply. For seller tools, measure time saved, mistakes and whether people keep using the tool after the trial.
Account for activity that would have happened anyway. A buyer switching from the site's search box to an AI assistant may enjoy a better experience, but every resulting sale is not automatically additional revenue. A control group, where practical, helps separate the two.
There is no need for a universal rule such as waiting until AI supplies 3% of visits. A small number of valuable enquiries may justify the cost. A large number of unproductive requests may not.
For gumtxxx and okxxxx, I would set a modest first goal: fewer pages for buyers to sift through, fewer irrelevant questions for sellers, and a better chance of matching available items with interested people.
If an MCP integration helps deliver that, there is a reason to keep investing. Negotiating and paying on the buyer's behalf can wait until the earlier steps work well.