Agent Surface & Observability

What is an MCP endpoint and does your site need one?

Q: What is an MCP endpoint and does your site need one?

An MCP endpoint is a structured interface, built to the Model Context Protocol, that lets an AI agent invoke your site's functionality directly — running a check, fetching live data, or performing an action — rather than just reading your pages. Where llms.txt and schema make your site readable to AI, an MCP endpoint makes it usable by AI. For a tool or audit product it means an assistant can run the tool on a user's behalf instead of linking out. Most content sites don't need one yet; sites offering a callable service increasingly will, because agentic workflows reward businesses an agent can act through, not just cite.

By Paul Masterson · Published 2026-06-07 · Last reviewed 2026-06-07 · 7 min read

Readable versus usable

Most of this site's framework is about making your content readable to AI: structured data, llms.txt, clean semantic HTML. The Agent Surface pillar is about something further along — making your site usable by AI. The difference is the difference between an assistant describing your tool and an assistant running it.

The Model Context Protocol (MCP) is the emerging standard for that second thing.

What MCP actually is

MCP is an open protocol that defines how an AI agent discovers and calls external tools and data sources in a consistent way. An MCP server exposes a set of capabilities — for example "check a URL's AI readiness" or "look up product availability" — that any MCP-aware assistant can invoke, with typed inputs and structured outputs.

Concretely, instead of a human visiting your site and clicking a button, an agent can call your endpoint, get a structured result, and fold it into whatever it's doing for its user.

Does your site need one?

Be honest about your shape:

Content / brochure sites — probably not yet. Your priority is being readable: schema, llms.txt, answer units. An MCP endpoint with nothing to invoke is theatre.
Tools, calculators, audit products — increasingly yes. If the value of your site is a function, an agent that can call that function for its user is a powerful new distribution channel.
Data providers — yes, trending toward essential. If people query your data, agents will want to as well.

The defensibility flip side

Exposing a callable surface means thinking about abuse: rate limiting, authentication where appropriate, scraper detection, and GDPR-compliant handling of anything personal that flows through. Agent accessibility and agentic defensibility are two sides of the same coin — see the framework's Agent Surface pillar.

Where this site is heading

In Phase 3, the AI Readiness audit on this site becomes MCP-callable: an agent will be able to run a readiness check on a user's URL directly. That's the canonical example — a tool whose value multiplies the moment agents can invoke it.

What to do this week

Decide whether your site's value is content or a function.
If it's content, finish the readability pillars first — don't skip ahead.
If it's a function, sketch the one or two capabilities an agent would call.
Plan rate limiting and data handling before you expose anything.

Agent Surface & Observability

Agentic defensibility — the flip side of being AI-accessible

Opening your site to AI agents creates new attack surface. Here's how to stay accessible to the agents you want while controlling cost, abuse, injection, and data-protection risk.
Agent Surface & Observability

AI-representation observability — watching how assistants describe you

Most organisations have no idea what AI assistants tell people about them. Here's how to measure it, track it over time, and correct misrepresentation at the source.
Discoverability & Structure

What is llms.txt and why your site needs one

A plain-English explanation of the llms.txt standard — what it is, what goes in it, and why it's becoming table stakes for AI discoverability.