Standard
An experimental browser API letting web pages register structured tools via document.modelContext.registerTool, so AI agents can invoke page actions directly.
Shopify added WebMCP checkout tools on September 28 that update payment and place an order once the buyer confirms. The spec is still a draft with no Firefox or Safari implementation, so tools belong on the open session behind a review step.
Reality
The DeepDeck benchmark runs one task twice, once with the WebMCP bundle installed and once with the tool list forced empty, freezing the model configuration in between and leaving correctness unscored unless you supply an answer you checked yourself.
Reality
document.modelContext.registerTool works in Chrome only when enable-webmcp-testing is on, and the demonstration page falls back to a third-party polyfill everywhere else. Its scoring tool runs entirely in the browser.
Reality
An InfoWorld column sorts web agents into three homes and argues that businesses with compliance exposure should build the on-site one first. Its security case rests on published guidance from Anthropic and OWASP.
Reality
Docs7 takes a Mintlify config, serves Markdown from the URLs agents already request, and counts which AI products fetch what, a metric Context7 itself says measures provider visits rather than readers.
Reality
The proposal, in origin trial from Chrome 149, lets a page register named tools with JSON schemas for their inputs, which hands web teams a second interface to scope and to answer for when an agent acts on a user's behalf.
Reality
The proposal swaps DOM inference for tools the page registers itself, which is sound engineering and also a standing maintenance bill for whoever owns the state machine. No shipping browser is named in the write-up.
Reality
A modelContext that registers tools correctly can still refuse addEventListener, and the throw lands after the page has painted. The repair is a bounded poll plus a banner that names the mechanism actually carrying the app.
Reality
Ten pre-registered WebMCP tools are live on Liquid storefronts with nothing installed. Four of them write, and the changelog says nothing about how you would know they ran.
Reality
The experimental category grades semantics, CLS, llms.txt and WebMCP, and it is explicitly not a ranking factor. The scoring rule means most sites can pass it without touching an agent.
Reality