Your Website Is Now an API for AI Agents

MCP endpoints, llms.txt and agent manifests are live on my sites today. What each rung of the architecture ladder can expose to the agents now acting for your customers.

This post is part of Beyond the Hybrid, my guide to the web architecture ladder, from legacy CMS to sites that own their own database.

For twenty years a website had one audience: a person with a browser. It now has a second one: software acting on that person’s behalf. Assistants that research, compare and increasingly transact do not want your homepage. They want your data in a form they can call.

What agents can use today

Three layers matter, from simplest to most capable:

  1. Readable pages and structured data. Full server-rendered HTML plus schema.org JSON-LD. The baseline every rung can reach.
  2. Published manifests. An llms.txt file that summarises the site for language models, and JSON files describing products or resources. This site, OneFront, ClipMaster, Lakewood Ranch AI and Florida Man AI all publish an llms.txt.
  3. Callable tools. Endpoints an agent can query directly. This site exposes WebMCP-style REST tools for listing products and creating a checkout link. Lakewood Ranch AI and OneFront run live MCP (Model Context Protocol) endpoints.

What each rung can expose

Theme-level hybrid: excellent pages and structured data, plus static manifests. Callable tools are possible through custom REST routes, but they live inside WordPress’s request cycle and plugin stack.

Decoupled hybrid: the API already exists because the front end uses it. Turning parts of it into agent tools is a natural extension. That is how this site’s product and checkout tools work.

Full-stack: the data model is yours, so the agent interface can mirror the business exactly. Lakewood Ranch AI’s MCP endpoint exposes three tools, search_businesses, get_business and list_categories, that answer from the same records the website shows, because there is no CMS translation layer in between.

Why this changes the rebuild decision

Until recently, the case for leaving WordPress was mostly about speed and flexibility. The agent layer adds a new one: the business whose data is cleanly callable is the one an assistant can recommend with confidence, book with and buy from. That does not require going full-stack on day one. It does require knowing which rung can expose what your customers’ agents will ask for.

Where to start

  • Make sure every public page carries its content in HTML (see AI crawlers don’t run JavaScript).
  • Publish an llms.txt and accurate JSON-LD.
  • Pick one high-value question an agent might ask about your business, and expose an endpoint that answers it.

Where to go from here

If you are trying to work out which rung your site belongs on, start with the full ladder, or see rebuild options and price ranges. If you would rather just ask, tell me what your site has to do and I will tell you what I would build.