What is MCP in one paragraph?

Model Context Protocol: an open standard that lets you expose tools (functions) and data to AI assistants over a server they connect to. Instead of scraping your pages, a supporting assistant calls your search or availability tool and gets structured answers. Anthropic introduced it in 2024; adoption spread across major platforms through 2025.

Is leaving /products.json open on Shopify safe?

It exposes the same public catalog data as your product pages, in cleaner form. Closing it does not hide anything a scraper wants; it mostly closes the easiest legitimate integration path. Check how yours responds and make it a deliberate choice.

Do I need a feed, OpenAPI, and MCP?

No. One feed moves you from warn to pass on this check and is typically an afternoon of work. Add OpenAPI if you already run a public API, and consider MCP when agent-driven business justifies real integration effort.

Agent endpoints: feeds, OpenAPI, and MCP

Updated 2026-07-13 · 4 min read · trust pillar

Everything else in these guides makes scraping your site work better. Agent endpoints make scraping unnecessary: a product feed, an OpenAPI description, or an MCP server gives agents a machine-readable surface with zero parsing ambiguity. The agent asks; your endpoint answers with data instead of markup.

As of mid-2026 this is still early-adopter territory, which is precisely the opportunity. Most sites expose none of it, so the ones that do are the easiest vendors for agent builders and assistants to integrate, cite, and transact with. Our scanner treats this as upside only: having any endpoint earns the points, having none is a warning rather than a failure.

Why connectable beats scrapable

A feed keeps your prices and stock current in the agent's context; scraped copies go stale the day your catalog changes. A queryable endpoint lets an assistant answer is the blue one in stock in medium with certainty, and certainty is what turns a mention into a recommendation. When an agent can either parse a competitor's HTML or call your endpoint, your data arrives complete and correct, and theirs arrives as a best-effort guess.

The emerging agent-checkout programs point the same direction: as of mid-2026, assistant shopping integrations are onboarded through platforms and structured product data, not through better scraping. Feeds and schema are the prerequisite layer for all of them.

The menu, cheapest first

Shopify stores usually already expose a JSON catalog at /products.json and a cart API at /cart.js; check that yours responds rather than being closed off by configuration. WooCommerce exposes a REST index under /wp-json/. If you are on either platform, your endpoint work may already be done.

On any stack, the cheapest universal option is a feed: JSON Feed, RSS, or Atom listing your products or content, declared with a link tag so it is discoverable from your homepage. If you run a public API, publish its OpenAPI description at a predictable path such as /.well-known/openapi.json so agents can construct calls against it.

html

<!-- In <head>: make the feed discoverable -->
<link rel="alternate" type="application/feed+json"
      title="Acme products" href="https://acme.example/feed.json">

A minimal JSON Feed for a catalog

Titles and content_text should carry the facts an agent needs inline: price, availability, the essentials. One static file regenerated on catalog change is enough; it does not need to be a live API to be useful.

json

{
  "version": "https://jsonfeed.org/version/1.1",
  "title": "Acme Outdoor Gear: products",
  "home_page_url": "https://acme.example",
  "feed_url": "https://acme.example/feed.json",
  "items": [
    {
      "id": "https://acme.example/products/trailhead-2",
      "url": "https://acme.example/products/trailhead-2",
      "title": "Trailhead 2 Ultralight Tent, $349, in stock",
      "content_text": "Two-person three-season tent, 1.1 kg. Free US shipping. 60-day returns."
    }
  ]
}

MCP and the deeper integrations

MCP (Model Context Protocol) is an open standard, introduced by Anthropic in late 2024 and since adopted across the major assistant platforms, for exposing tools and data that models can call directly. A remote MCP server for a store might expose search-catalog and check-availability tools; assistants that support MCP can then work with your inventory as a first-class integration instead of a scrape. It is real engineering work, so do the feed first, but it is the strongest form of agent readiness available today.

For completeness: /.well-known/ai-plugin.json is the manifest format from the earlier ChatGPT plugins era. It is deprecated as of mid-2026 but still probed by tooling, including ours, as a sign a site thought about agent access. Do not start there; a feed or MCP server is worth more.

How AgentReady checks it

The scanner asks: are emerging agent endpoints exposed? Any one of these passes: a reachable /products.json, an RSS, Atom, or JSON Feed declared via link tag, an OpenAPI description, an ai-plugin.json manifest, or MCP endpoint hints. Exposing none warns rather than fails, because this is an early-adopter check. It is worth 4 points in the Trust pillar.

The methodology's reasoning: feeds, plugin manifests, and agent endpoints are how the newest agents connect directly. Exposing any of them puts you ahead of sites they have to scrape.

Frequently asked questions

What is MCP in one paragraph?: Model Context Protocol: an open standard that lets you expose tools (functions) and data to AI assistants over a server they connect to. Instead of scraping your pages, a supporting assistant calls your search or availability tool and gets structured answers. Anthropic introduced it in 2024; adoption spread across major platforms through 2025.
Is leaving /products.json open on Shopify safe?: It exposes the same public catalog data as your product pages, in cleaner form. Closing it does not hide anything a scraper wants; it mostly closes the easiest legitimate integration path. Check how yours responds and make it a deliberate choice.
Do I need a feed, OpenAPI, and MCP?: No. One feed moves you from warn to pass on this check and is typically an afternoon of work. Add OpenAPI if you already run a public API, and consider MCP when agent-driven business justifies real integration effort.