Somewhere this week, a buyer asked an AI assistant to find them three suppliers, compare them, and draft the enquiry email. The assistant visited your website to do it. Whether it found anything it could use is a different question.
That visitor is an agent: software reading on a buyer's behalf. It doesn't see your design, your animations, or the hero video. It parses your text, your structure, and your data, and it recommends based on what it can extract. Here's what agents can and can't read, where MCP fits, and what's worth doing about it this quarter.
What an agent sees when it reads your site
Strip a webpage of its visuals and you get what the agent gets: headings, text, links, and whatever structured data sits in the markup. A site that looks impressive but speaks in vague claims parses as nothing. "Industry-leading services for your business needs" extracts to zero facts. "Structural steel fabrication, Melbourne, 10 to 400 tonne projects, typical lead time six weeks" extracts to five. The agent isn't judging your brand. It's harvesting claims it can repeat, and it can only repeat what you've stated plainly.
The four layers that make a site agent-readable
In order of importance. Semantic markup: real headings, real lists, content in the HTML rather than rendered by scripts the agent may not run. Structured data: schema that states your services, prices, locations, and FAQs as machine-readable fact. Answer-first content: pages that state the answer before the argument, written so a paragraph survives being lifted out of context. And published pricing: an agent comparing suppliers can't book a discovery call, so the sites with numbers get shortlisted and the sites with "contact us for a quote" get skipped.
Where MCP fits, and where it doesn't
MCP (Model Context Protocol) is the open standard that lets an agent do more than read: query your availability, check a price, place an order, book a time. It's the difference between a site an agent can quote and a site an agent can transact with. We built one for our own site and wrote up what happened. The short version: it won't move your AI visibility today, and for most businesses the transactional layer is early. Readable comes first. If agents can't extract what you do and what it costs, an interaction layer on top has nothing to work with.
What to do this quarter
Run your key pages through the strip test: text only, no visuals, and ask whether a stranger could work out what you do, for whom, at what price. Fix the pages that fail with plain answers to the questions buyers ask before they buy. Add FAQ and service schema where it's missing. Publish your prices as ranges if exact numbers don't exist. And keep your business details identical everywhere they appear, because agents cross-check, and inconsistency reads as unreliability.
Most businesses don't need an MCP server yet. Every business needs a site a machine can read, because the machines are already reading. If you want to know what the agents find when they look at yours, that's the first thing we check. Tell us what's going on. We'll tell you what we'd do and what it costs.
FAQ
What is an AI agent in the context of buying?
An AI agent is software that acts on a person's behalf: researching suppliers, comparing options, shortlisting, and increasingly transacting. When a buyer asks an assistant to find and compare providers, the agent visits websites, extracts what it can read, and recommends based on that.
What makes a website agent-readable?
Content in real HTML rather than script-rendered, structured data stating services, prices, and locations, answer-first pages that survive being quoted out of context, and published pricing. Agents shortlist sites they can extract facts from and skip sites that only make claims.
What is MCP?
MCP (Model Context Protocol) is an open standard that lets AI agents interact with a business's systems: checking availability, querying prices, placing orders. It goes beyond reading a website to transacting with it.
Does my business need an MCP server?
Probably not yet. For most businesses in 2026 the priority is being readable: clear content, structured data, published pricing. MCP matters once agents in your category start transacting, and building it before your site is readable puts the layers in the wrong order.
Why does published pricing matter to AI agents?
An agent comparing suppliers can't sit through a sales call. Sites that publish numbers give the agent a fact to compare and repeat. Sites that hide pricing give it nothing, and often get dropped from the shortlist.
