Skip to main content

WebMCP and the Agent-Ready Web

For thirty years the web assumed a person was reading it. That assumption is quietly ending. Software now shops, compares and books on someone's behalf, and a website built entirely for human eyes is close to useless to it. WebMCP is the standard that changes what a site can offer an agent, and it is worth understanding before it matters to your revenue.

August 27, 20267 min read

What WebMCP actually is

WebMCP is a proposed browser standard that lets a page hand an AI agent a set of tools. The page calls a browser API, navigator.modelContext, and registers named actions with descriptions and typed inputs. An agent operating that browser reads the list and calls what it needs.

The change is subtle and large at once. Today an agent asked to book a consultation has to behave like a person: find the button, guess the form, hope the calendar widget is operable. With WebMCP the site says what it can do, and the agent calls the action directly. Guesswork becomes an API call, and the failure modes that make agent browsing unreliable mostly disappear.

It is the in-page counterpart to the Model Context Protocol, the standard assistants already use to reach external systems. MCP serves an agent anywhere. WebMCP serves an agent that is already on your page. Support is arriving first in the agent-capable browsers and assistant clients, which is why the surface is cheap to claim right now.

The three layers of being agent-ready

  • Readable

    An assistant summarizing your business needs facts it can lift without guessing: what you sell, what it costs, who you serve. That is llms.txt, clean structured data, and prices that are not trapped behind a form.

  • Discoverable

    An agent arriving cold needs to learn what you can do in one fetch. An A2A Agent Card at a well-known path is the business card: your skills, your endpoints, your contact route, as JSON.

  • Invokable

    Reading is not doing. An MCP server exposes your booking calendar or catalog as tools an assistant can actually call, and WebMCP does the same thing inside the page itself for agents driving a browser.

Most sites are at zero of three. That is not a criticism, it is the current state of the web, and it is exactly why the work is inexpensive today and will not be once it is table stakes.

Why a normal website fails an agent

Ask an assistant to compare three local firms and it will do its best with visible text. It can summarize your services. What it usually cannot do is answer the question that decides the shortlist, which is what you charge. Pricing that lives behind a form is invisible to a machine, so the assistant does the reasonable thing and recommends whoever published a number.

This is the part worth sitting with. Being unreadable does not put you lower on the list. It takes you off the list, because the agent has nothing to compare. That is why our own prices are published in full on the pricing page and served as a machine-readable feed, and why we treat published pricing as a competitive position rather than a disclosure risk.

We did this to our own site first

Everything described above is live here, and you can check it without asking us. Fetch our Agent Card and you get our skills and endpoints as JSON. Point any MCP-capable assistant at /api/mcp and it gains six tools over our catalog, pricing and case studies. Every page of this site registers its skills in-page through WebMCP. The full policy, including what happens when an agent asks for something we have not built, is on the agents page.

One rule governs that layer: nothing is advertised that is not implemented. Anything outside the published skills gets a structured redirect to a monitored human inbox rather than an invented answer. We would hold a vendor to that standard, so we hold ourselves to it.

The service, and what it costs

Agent-Ready Essentials is $599 setup plus $300 monthly. Your business made readable and invokable by AI agents: an A2A Agent Card at your well-known path, a machine-readable offers and services feed, llms.txt files built or repaired, a read endpoint over your business facts, and honest out-of-scope routing to your own inbox. Monthly covers protocol updates and an accuracy verification against your live prices.

Agent-Ready Pro is from $2,500 setup plus $300/mo. An MCP server over your actual systems: your booking calendar, your catalog or inventory, your CRM read paths, exposed as tools any MCP-capable assistant can use. Scope-quoted, and the natural bridge into a Work AIOS System.

If you would rather start by finding out where you actually stand, the free AI Automation Audit now includes an agent-readiness check, and the Agent-Ready service page lists exactly what ships. For how this fits the wider question of getting recommended by assistants, see the guide to AI search visibility.

WebMCP and agent-readiness questions

What is WebMCP?+

WebMCP is a proposed web standard that lets a page register tools directly with an AI agent through a browser API called navigator.modelContext. Instead of an agent guessing which button to click, the site declares what it can do and the agent calls it. It is the in-page sibling of the Model Context Protocol, which does the same job server side.

How is WebMCP different from an MCP server?+

An MCP server lives at a URL and any MCP-capable assistant can connect to it, whether or not anyone is browsing your site. WebMCP runs inside the page and serves an agent that is already there, driving a browser on someone's behalf. Most businesses eventually want both, because they serve two different arrival paths.

Is this worth doing before agent traffic is large?+

That is the honest question, and the answer is that the work is small now and the position compounds. Publishing structured facts and a machine-readable price list also improves how assistants describe you today, which is measurable in AI referrals well before anyone's agent is buying. If an assistant cannot state your price, it recommends someone whose price it can state.

What does an agent see if we do nothing?+

A page built for human eyes. It can read visible text, so a summary is possible, but anything behind a form, a calendar widget, or a Contact us for pricing button is invisible to it. In practice you become the option the agent cannot evaluate, which is functionally the same as not being on the list.

How do we check whether our site is agent-ready?+

Fetch three things and see what comes back: your-domain.com/llms.txt, your-domain.com/.well-known/agent-card.json, and your robots.txt. Most sites return two 404s and a robots file that has never been told what to do about AI crawlers. That is the starting position for nearly everyone right now.