What MCP is, in one paragraph
The Model Context Protocol is an open standard, first published by Anthropic in November 2024, for connecting AI applications to outside tools and data. An MCP server describes what it can do (tools the model can call, resources it can read, prompt templates) in a machine-readable way, and any MCP client, such as Claude, Cursor, VS Code or ChatGPT, can connect to it without custom glue code. Messages are JSON-RPC 2.0, sent over standard input and output for a local server or over HTTP for a remote one.
What WebMCP is
WebMCP is a proposal for the web platform, not a server you install. A site registers tools with the browser (the explainer uses a navigator.modelContext API, and a declarative form where an HTML form itself is the tool), each with a name, a description and an input schema. When an agent that lives in the browser wants to act on that site, it reads the registered tools and calls them, with the user's own session, instead of guessing at buttons and inputs in the DOM.
The pitch is reliability and consent. Today a browser agent clicking through a checkout breaks whenever the layout changes; with WebMCP the site says exactly what can be done and with which fields, and the user stays signed in as themselves. The name is deliberate: the tool shape (name, description, JSON Schema) is borrowed from MCP so the same models and the same client code can use both.
The difference that matters
MCP connects an agent to things that are not the page it is looking at: your database, your mail, a SaaS API, a file system. The server runs somewhere, speaks JSON-RPC, and any client can connect. WebMCP connects an agent to the page it is already on. Nothing runs outside the browser and nothing is reachable from a terminal agent like Claude Code or Codex.
So they are not competitors for the same job. A company might publish an MCP server for developers and automations, and add WebMCP tools to its web app for people who browse with an agent. A reader choosing between them is usually asking a different question: do I want an agent to use this service from anywhere (MCP), or on this site while I watch (WebMCP)?
One practical difference today is maturity. MCP has clients, servers, a registry and a year of security incidents to learn from. WebMCP is an explainer and experimental builds; the API may change before it ships broadly.
What to check before trusting either
For an MCP server: who published it, whether it runs code on your machine or only talks to an endpoint, which tools write or delete, and whether the tool descriptions say anything a tool has no reason to say. That is what the grade on every listing here measures, with the evidence shown.
For WebMCP: the tools come from the site you are on, so the trust question is the site's. Watch what the page registers and what the agent is allowed to do with your logged-in session. The same prompt-injection risk applies: a page can describe a tool misleadingly to steer the agent, so an agent should treat tool descriptions as data from the site, not instructions from you.