For years, we’ve built websites primarily for people. We invest in design, navigation, content, search engine optimisation and increasingly sophisticated ways of encouraging visitors to take action. But what happens when the visitor is no longer a person, but an AI Agent acting on their behalf?

I’ve been thinking about this recently while adding an MCP (Model Context Protocol) interface to my own AI4Enterprise website. Rather than having an AI Agent browse the website, interpret pages and hopefully find what it needs, why not give it a way to ask the website directly?

I think we may be looking at the beginning of another significant change in how the web works. Not the end of websites, but potentially a fundamental change in who, or what, they are designed to serve, which has some interesting implications for businesses.

What is it?

Today, much of the web seems to operate around a fairly familiar model (I’ll be honest - it’s not my field of expertise). A business publishes information on a website, search engines index that information then people search, find the website, navigate through pages, compare information and eventually take some action.

AI is already changing the middle of that process. People increasingly ask AI assistants questions rather than searching through pages of results themselves. AI-powered browsers can navigate websites, collect information and, with the appropriate permissions, take actions.

But I think there is another step beyond that.

What if websites stopped being just collections of pages and started exposing their capabilities directly to AI Agents?

Imagine I’m looking for an AI consultancy. Rather than searching Google, reviewing websites and reading about services, I ask my AI Agent to find three suitable companies with experience in my sector, relevant qualifications, demonstrated delivery experience and availability for an initial discussion. The Agent could potentially interact directly with each company’s systems, asking for its services, credentials, experience, sectors served and ways to get in touch.

No browsing through an About page. No interpreting marketing material spread across multiple pages. The Agent asks questions and receives structured responses that it can evaluate.

This is the Agentic Web that’s becoming clearer to me (and this is just one evolutionary step).

A small experiment on my own website

I’ve recently added an MCP server to my AI4Enterprise website. It’s a pretty straightforward idea, rather than expecting an AI Agent to browse my website, I’ve exposed a series of tools that allow it to ask questions about the business. It can search published articles, retrieve their contents, examine services, obtain company information and find contact details.

An Agent can ask what AI4Enterprise does, what industries it serves, what evidence supports its expertise or what I’ve written about a particular subject. The website is still there for people, but there is now another interface intended for machines.

The human vs. machine view of the website is important - an MCP server isn’t simply another chatbot on the website. It exposes capabilities that compatible AI systems can use as part of their own workflows, once they have discovered or been configured to access the endpoint.

Back in February 2025, I wrote an article asking “Are AI Agents the New APIs?“. My thinking then was that agents could fundamentally change how applications communicate and integrate with one another. APIs would remain important, but agents could introduce a more flexible, intent-driven layer above them (I think that’s the next iteration of this evolution and an AI Agent might be next on my list accessible at /agent over A2A). This feels like a continuation of the thinking behind that earlier article, only now applied to the public web.

And now for the really interesting question: how would an Agent even know that a website has an MCP interface? Today, MCP servers can expose an HTTP endpoint, but there isn’t a universal requirement for every website to publish it at a particular address. What if the future standard were as simple as appending /mcp to a website’s URL?

In my case it’s: https://www.ai4enterprise.ca/mcp - simple.

An Agent looking at a website could check that predictable address, discover the capabilities offered and decide whether they are useful. That’s a possible future convention, not an established universal standard.

The underlying idea seems compelling. Just as websites became discoverable and searchable by machines, why shouldn’t their services and capabilities could become discoverable and usable by Agents.

What does it mean from a business perspective?

I think the implications could be considerably broader than adding another technical interface to a website.

The customer journey could change.

For years it seems we’ve invested in getting people to visit our websites. SEO, digital advertising, landing pages, content marketing, conversion rates and analytics are all built around that model (and I don’t think SEO or the bulk of this changes - just the way we interact).

But if a prospective customer’s AI Agent researches suppliers, compares offerings and makes recommendations without the customer ever visiting the websites, where does that leave the traditional model? The Agent will become the first point of contact between an organisation and its prospective customers.

And that raises some interesting considerations.

1. Discovery and visibility could change.

Getting onto the first page of Google seems to have been an important commercial objective. In an Agentic Web, the more interesting question might be whether an AI Agent can find, understand and accurately represent your organisation when evaluating options for its user.

SEO won’t disappear (maybe it will become even more important), but being discoverable by agents, along with being agent-interaction friendly, could become another important consideration.

2. The quality of information becomes even more important.

An Agent evaluating potential suppliers needs more than attractive marketing language. It needs reliable information about services, capabilities, evidence, experience, locations, qualifications and maybe even commercial terms. And that information needs to be current, consistent and verifiable.

An organisation with a beautifully designed website but poor underlying information may be at a disadvantage compared with one that makes its capabilities easy for AI systems to understand.

3. Websites could become service interfaces.

The initial use case might simply be retrieving information, but why stop there? An organisation could potentially expose capabilities that allow an authorised Agent to check product availability, request a quotation, book an appointment, submit an application or initiate a transaction (whether this is MCP or A2A is up for debate).

This is where MCP, existing APIs and potentially Agent-to-Agent (A2A) communication could start coming together. The website becomes more than somewhere to read about services. It becomes a controlled way for external Agents to interact with the organisation.

4. Trust, identity and governance become critical.

Of course, there’s a huge difference between letting an Agent read your published articles and allowing it to perform a financial transaction. Who is the Agent acting for? How do we establish its identity and authority? What information can it access? What actions can it perform? How do we manage security, privacy, prompt injection, auditability and accountability?

Opening up systems to Agents without thinking through these issues would be a mistake. I wouldn’t advocate making every capability available through MCP simply because we can. Existing APIs, authentication, authorisation and good integration architecture remain essential.

5. We may lose some control over the customer experience.

This is perhaps one of the more interesting commercial consequences. If an AI Agent is the interface between your company and the prospective customer, you may no longer control how your organisation is presented.

The Agent decides what information to retrieve, how to compare it and what to include in its recommendation. Your branding, website design and carefully constructed customer journey may never be seen - that remains in the human domain.

That doesn’t mean branding stops mattering. It means organisations may need to think about how their value and differentiation are represented when an AI system becomes the intermediary.

What do I do with it?

I don’t think businesses need to rush out and replace their websites with MCP servers, or even add MCP servers to their website (my website is an extension of my experiementation). We’re still at an early stage, and the standards, tools and commercial models will continue to evolve.

But I do think this is worth starting to consider, particularly for organisations investing in new websites, digital services or broader AI strategies.

A few practical starting points:

• Try looking at your business through an Agent’s eyes. Ask an AI system to research your organisation, compare it with competitors and recommend whether someone should do business with you. What does it find? What does it miss? What does it get wrong?

• Get your information in order. Make sure services, capabilities, credentials, evidence and contact information are accurate, current and consistent. This matters regardless of which agent protocols eventually prevail.

• Explore machine-accessible interfaces. If you already have APIs, structured data or other services, consider whether a read-only MCP interface could make some of those capabilities easier for Agents to use.

• Think about discoverability. How would an Agent know your capabilities exist? What would it need to understand them? This may become another consideration for digital architecture and marketing teams.

• Separate information from action. Public information retrieval is one thing. Allowing an Agent to make commitments or change business data is another. Establish appropriate security, permissions and governance before exposing transactional capabilities.

• Experiment before committing. My own MCP website experiment is an example of something relatively small that can help explore these questions without redesigning the entire digital presence (it’s a relatively easy bolt-on - much of the work was in making sure the appropriate data was exposed for the test-case I used).

I also think there is an interesting architectural question here. We’ve spent decades separating presentation layers from application and data layers. That separation may become even more valuable if the presentation layer is no longer always a human-facing user interface.

A company website will have two, possibly three quite different front doors: one designed for people and others designed for AI Agents (one via MCP and another via A2A accessible via /mcp and /agent respectively).

Both accessing the same underlying business capabilities, but in very different ways. Perhaps that leads to a more fundamental change in how we think about digital presence. For years, the objective has largely been to make it easier for people to find and interact with our organisations online.

In the next phase, we may also need to make it easier for their AI Agents to find, understand and interact with us on their behalf. The web doesn’t disappear, neither do websites, APIs or search engines - but the audience changes.

And if that happens at scale, I wonder whether we will look back at today’s websites in much the same way we now look at the early web, as an important step in the evolution of how people and organisations interact digitally.

The question for business leaders may no longer be just “How do we get people to visit our website?”

It may increasingly become:

“How do we make sure their AI Agents can find us, understand what we do and work with us?”

That’s quite a different way of thinking about your digital strategy.