Field note

Your website is becoming something AI talks to, not just reads

A 2026 platform report logged 102 million Model Context Protocol sessions in a single year. A sign websites are turning into surfaces AI agents call, not just pages humans read. Here's what the data shows, what it doesn't, and how to get ready for both.

Buffy Editorial2026-06-26 · 4 min read

Your website is quietly becoming a surface AI agents act on, not just a page humans read. A 2026 platform report logged about 102 million Model Context Protocol sessions in a single year. Early but real evidence that AI-to-website interaction is turning into routine infrastructure. The durable shift: a site now has two audiences, the human visitor and the AI agent, and the work of being legible to both is increasingly the same work.

The data here comes from Wix's 2026 State of Websites report, drawn from its base of about 310 million users. A single-vendor dataset, so the figures describe one platform's customers, not the whole web. Read them as directional.

What did the MCP-adoption data actually show?

It showed AI managing websites at meaningful scale. MCP. The Model Context Protocol, an open standard that lets AI assistants connect to external tools and data through a consistent interface. Moved from novelty to routine on Wix in 2025:

Metric Figure What it measures
Total Wix MCP sessions in 2025 ~102 million AI assistants connecting to Wix sites
Users connecting external AI tools to their site ~32,000 Site owners wiring up an AI like Claude
Sample ~310 million Wix users Scope of the report (single platform)

Wix describes site owners using MCP to create and update content, search documentation, and debug issues, and external assistants like Claude managing a site "through natural conversation." Be precise about what this is: most of these sessions are owners using AI to manage their own sites, not consumer shopping agents transacting on them. That distinction matters, because it's tempting to read 102 million sessions as "agents are buying". They aren't, yet. What the number does signal is that the plumbing for AI-to-website interaction is being laid, fast.

Why does this matter for AI visibility?

Because it confirms the direction every agentic protocol is pointing: the website is becoming a two-sided interface. For years a page had one job. Be read by a human (and, lately, crawled by an AI). Now a second job is emerging: be callable by an agent that wants to do something. Answer a question, pull a spec, start a cart.

A website used to be a destination you sent people to. It's becoming an endpoint that machines query, and the brands that win will be legible to both at once.

This is the same theme as making your website agent-readable, now showing up in usage data rather than guidance. It also sits alongside the consumer-side agentic commerce protocols we've covered: WebMCP vs UCP on how browser agents and commerce agents reach a site, and Microsoft's Web IQ on grounding agents in live web data. The Wix figures are the owner-side mirror of that consumer-side shift. Both say the same thing: AI is becoming an active party in how websites are built, read, and acted on.

What does the data not tell you?

It doesn't tell you that you need a separate "AI version" of your site, and it doesn't tell you consumer agents are transacting at scale yet. Two honest limits:

  • It's owner-side, not shopper-side. The MCP sessions are largely people managing their own sites with AI. Useful as a signal of momentum, not proof that buying agents have arrived.
  • It's single-vendor. The figures come from one platform's customers. They show a trend on Wix; they don't quantify the open web.

The trap to avoid is over-reacting by building a stripped-down machine mirror of your site. A move we've argued against in whether you need a separate AI version of your website. A divergent mirror risks cloaking and drift. The better response is to make your real site legible to machines, not to build a second one.

How should a brand get ready?

Invest in the fundamentals that serve both audiences, and watch the protocol layer without betting the roadmap on it:

  • Server-rendered, semantic HTML. If a fact only appears after JavaScript executes, an agent may never see it. Expose it in the markup.
  • Clean structured data. Schema labels what things are so an agent doesn't have to guess; it's the cheapest agent-legibility win available.
  • Reachable by AI bots. Confirm your CDN isn't blocking AI crawlers and that your pages are in the sitemap and internally linked.
  • Stay informed on protocols, experiment lightly. WebMCP, UCP, and MCP are moving fast; track them and pilot, but don't pour budget into one standard while they're still settling. The same caution applies to llms.txt. A control you fully own is a weak signal.

The fundamentals are durable; the protocols are volatile. Do the legibility work now, because it pays off whether an agent reaches your site through search, a browser assistant, or a commerce protocol. Knowing whether AI agents and answer engines actually reach, read, and recommend your site. Across every one of those paths. Is exactly what Buffy Intel tracks. Questions: [email protected].

Frequently asked

What is the Model Context Protocol (MCP)?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data through a consistent interface, so an assistant like Claude can read from, or act on, a connected system through natural conversation instead of a custom integration. In a website context it works two ways: site owners use it to let an AI manage their site, and, increasingly, it points toward websites exposing capabilities that external agents can call. Treat it as an emerging, fast-moving standard.

Does the Wix MCP data mean shopping agents are buying through websites now?

Not directly, and it's important to be precise. Wix's 2026 State of Websites report logged about 102 million MCP sessions in 2025 and ~32,000 users connecting external AI tools to their sites, but those numbers mostly measure site owners using AI to build and manage their own sites, not consumer shopping agents transacting. The honest read is directional: it shows AI-to-website interaction becoming routine infrastructure, which is the same direction consumer agent protocols are heading.

What should I actually do to prepare my site for AI agents?

Start with the fundamentals that serve both humans and machines: server-rendered, semantic HTML; clean structured data; a crawlable, well-linked site that AI bots can reach. Then watch the agent-facing protocols. WebMCP on the browser side and UCP for commerce. Without over-investing in any single one while they're still settling. The durable work is making your facts machine-legible; the protocol layer is where you stay informed and experiment, not bet the roadmap.