WebMCP vs MCP
MCP (Model Context Protocol) lets an AI app use tools on a server, any time, with its own sign-in. WebMCP lets a web page offer tools to an AI agent working in the browser, only while the page is open, using the visitor's own session. MCP is for background work; WebMCP is for doing things on a site the person is looking at.
By Jude Eric ·
Side by side
| MCP | WebMCP | |
|---|---|---|
| Where tools run | On an MCP server, local or remote | In the web page's JavaScript, in the visitor's browser tab |
| Who uses them | An AI app (Claude, ChatGPT, Gemini, Muse…) | An AI agent operating in the browser |
| When they exist | Whenever the server is up | Only while the page is open |
| Sign-in | The server's own (OAuth for remote servers) | The visitor's existing browser session |
| Needs a person there? | No: can run in the background | Designed for a person with the page open |
| Offers | Tools, resources and prompts | Tools only |
| Status | Widely supported by AI apps | A proposed browser standard; origin trials in Chrome (from Chrome 149) and Edge |
Checked 14 November 2026.
Where WebMCP stands in November 2026
- Not a finished standard. The spec is a W3C Community Group draft (20 October 2026), not on the W3C Standards Track. The proposal's authors are from Google and Microsoft.
- Chrome: an origin trial from Chrome 149, plus a testing flag (
chrome://flags/#enable-webmcp-testing). It's not in stable Chrome for everyone. - Edge: an origin trial that runs until 30 March 2027.
- The API moved. Chromium removed
navigator.modelContexton 9 July 2026; new code usesdocument.modelContext.registerTool(). Many older guides still show the old name. - HTML-only tools (marking up a form so agents can use it) are described in Chrome's docs but aren't in the draft spec yet.
Which one should you build?
- Your service has data or actions an AI should reach any time (bookings, orders, account lookups): an MCP server.
- People use AI agents on your website and you want those agents to fill forms and click through flows reliably: WebMCP.
- Both, usually: MCP as the core service, and WebMCP for actions on your pages while a visitor is there. Chrome's own guidance suggests the same split.
How YeNo uses both
A YeNo public business page has both: a read-only MCP server at its own address, so any AI can look up the business's hours, prices and policies, and, when the owner turns requests on, a booking form marked up for WebMCP so an agent in the browser can send a request. The same facts are also a small Markdown file at /b/<name>.md (and /llms.txt) for AIs that can only read a link. Either way, nothing is confirmed until the owner says so on their phone. YeNo for business
Questions, answered
01What is the difference between WebMCP and MCP?
MCP tools run on a server and an AI app can use them any time with the server's own sign-in. WebMCP tools live in a web page and only exist while the page is open, using the visitor's own browser session.
02What is WebMCP?
WebMCP is a proposed browser standard that lets a web page offer tools, with a name, a description and inputs, to AI agents working in the browser. In November 2026 it's in origin trials in Chrome and Edge.
03Does WebMCP replace MCP?
No. They solve different problems: MCP for background work and data access, WebMCP for actions on a page a person has open. Many sites will use both.
04Is navigator.modelContext still used?
No. Chromium removed navigator.modelContext on 9 July 2026. Current code uses document.modelContext.registerTool().
Cite this page
Jude Eric, YeNo Rules. “WebMCP vs MCP.” Last updated . www.yenorules.com/guides/webmcp-vs-mcp