Most of what a website communicates is built for a human to look at: a hero image, a headline, a grid of service cards with icons. An AI agent trying to answer a specific question — "what services does this company offer, and how do I contact them?" — has to scrape that page and guess at the structure. WebMCP removes the guessing.
What WebMCP actually is
WebMCP is a lightweight convention for exposing a small set of structured endpoints — a discovery file plus a handful of API routes — that let an AI agent query specific facts about your business directly, in a predictable JSON format, instead of parsing rendered HTML.
In practice, it has three parts:
1. A discovery file
Published at a fixed, predictable location (/.well-known/mcp.json), this file tells an agent what your business is, how to contact you, and which structured endpoints are available to query. It is the entry point an agent looks for first.
2. Structured API endpoints
Each endpoint answers one specific type of question in a consistent format — for example, a get_services endpoint that returns your service list with descriptions, and a get_contact endpoint that returns your phone, email, and address as structured fields rather than text an agent has to interpret.
3. A reference in llms.txt
We link the discovery file from our llms.txt, so an agent that has already found that file also finds the MCP endpoint without needing to guess at a URL.
How we built it for InfinitusNow
Rather than only describe this in the abstract, we implemented it on our own site: our discovery file lives at infinitusnow.com/.well-known/mcp.json, and it points to two endpoints — one returning our service list, one returning our contact details — both plain JSON, no scraping required. It is a small implementation, deliberately: two endpoints cover the two questions an agent evaluating our business is most likely to ask.
Why this matters more than it might seem
Most businesses will never need to think about this the way they think about their website's visual design. But the businesses that have accurate, structured data available to agents will be evaluated correctly and consistently, while businesses that rely on an agent correctly scraping and interpreting a marketing webpage risk being misread, or skipped entirely because the information could not be confidently extracted.
It is a small, low-cost piece of infrastructure with an asymmetric payoff: cheap to build once, and it does not need to be maintained much beyond keeping the underlying data current.
Where this fits into a broader AI-visibility strategy
WebMCP is one piece of what we call AAO (Assistive Agent Optimisation) — see our full AAO guide for the complete picture, including llms.txt and structured schema, or our AAO playbook for the practical checklist.