Particle.news
Download on the App Store

WebMCP Proposal Reimagines Web Crawling for AI Agents

By exposing structured tool contracts, sites would let agents skip UI guessing to execute tasks directly.

Overview

  • An analysis argues that WebMCP would move agents from parsing pages to invoking site-defined actions such as searchFlights or checkout with explicit parameters.
  • It contrasts today’s screenshot- and DOM-driven loops—described as slow, token-heavy, and brittle—with direct function calls that bypass visual inference.
  • The piece cites reported evaluations claiming up to 89% token savings and more than 60% less processing compared with screenshot-based methods.
  • The approach is framed as a shift from a document web toward an action web where websites advertise capabilities rather than just content.
  • The article warns that sites without such capabilities could become less visible to agents, potentially altering how discovery and ranking work.