MCP and WebMCP

I’ve used MCP for a while but lately I’ve been reading up on the new WebMCP, and the names make the two sound like the same thing. They aren’t.

MCP

An agent connects to a server that you run.

Runs

As a separate program, on your machine or a server.

Setup

Build or install a server, add it to the agent’s settings and give it keys.

Used by

Agents you set up yourself. No browser has to be open.

Good for

Back-office work: catalogue, reports, files.

WebMCP

A web page offers its own tools to the browser.

Runs

Inside the page, in an open browser tab.

Setup

A few lines of JavaScript, or toolname and tooldescription on a form. Nothing new to host.

Used by

A visitor’s browser assistant, with their login, where they can watch.

Good for

What visitors do on a site: search, filter, book, order.

MCP needs someone to build a server, run it and look after its credentials. In return it works anywhere, with no browser open, which suits background jobs.

WebMCP has no server. The page registers a tool, and a browser assistant can call it instead of guessing its way through buttons and fields:

document.modelContext.registerTool({
  name: "search-products",
  description: "Search the catalogue by keyword.",
  inputSchema: {
    type: "object",
    properties: { query: { type: "string" } },
    required: ["query"],
  },
  execute: ({ query }) => searchCatalogue(query),
});

So they answer different questions. MCP: what can my own agents do with my systems? WebMCP: what can a visitor’s assistant do on my site?

I don’t see one replacing the other. The first thing I want to try with WebMCP is product search, where a wrong call costs nothing.