Scan a website

Learn

MCP vs WebMCP

Same tool-calling idea, two very different deployment models.

MCPAI applicationconnects to a separate MCP server
WebMCPAI agent, already on the pagecalls tools registered by that same page

Model Context Protocol (MCP)

MCP is a general, transport-level protocol for connecting an AI application to external tools and data sources — typically a separate server process (local or remote) that an AI client connects to explicitly. A business chooses to stand up an MCP server and publishes it for agents that know to look for it.

WebMCP

WebMCP applies the same tool-calling idea inside a normal web page. Instead of a separate server, the page itself registers tools in the browser at runtime via a JavaScript API. An agent that's already browsing the page — not a separately configured MCP client — can discover and call those tools.

The practical difference

  • Where it lives: MCP runs as its own server; WebMCP runs inside the website an agent is already visiting.
  • Discovery: MCP servers are typically configured explicitly ahead of time; WebMCP tools are (currently) only discoverable by visiting the page.
  • Maturity: MCP is an established, widely adopted protocol. WebMCP is an early-stage browser proposal, currently limited to a Chrome origin trial.

Why MCPWebsite covers both

Most websites today aren't running either. Our score reflects a broader question — general AI Agent Readiness — of which WebMCP support is one forward-looking signal among several, not the whole picture.