The W3C’s new WebMCP project could change how AI agents interact with digital services. But who controls what happens when machines start acting on our behalf?
On 27 October 2026, the World Wide Web Consortium (W3C)
announced an experimental project supporting WebMCP, a proposed interface that allows websites to expose their functions directly to AI agents. Instead of scraping the site via links and traditional interactions (something that has been driving up the hosting costs of large volume websites for months), the new project aims to allow developers to plan and implement distinct interactions for humans and AI.
The creative perspective“WebMCP has the potential to fundamentally change how we approach designing websites. We’re now looking beyond how human beings navigate pages and achieve their goals into how to make it simultaneously easier for AI agents to complete their tasks. The challenge will be to do this while maintaining the brand experience, building trust and – most importantly – retaining human control.”
Mark Gristock, Ideation Director, Psycle
The technical perspective“We need to be 100% sure that the requests coming in via MCP are subject to the same rules and regulations as web requests. To that end, integration with frameworks will be complicated – but not impossible. We need to ensure a user using MCP can’t do more (or less) than they’re supposed to.”
James Partington, Technical Director, PsycleThe takeawayFuture websites may need to support both human interaction and structured, permission-controlled machine interaction. We need to start thinking about how to build websites that support humans arriving at different stages of the conversion funnel, and which elements could be completed in external AI agents for later benefit.
A question for youIf an AI agent could complete tasks on your website tomorrow, which would you allow it to perform independently, and which would still need human approval?