Current AI agents must simulate human behavior to use websites, which is inefficient and gives bots too much power.
WebMCP is a protocol that allows websites to expose functions, or tools, for agents to use programmatically via declarative or imperative APIs.
A live demo shows how to enable WebMCP in Chrome and register a JavaScript function as a tool that an agent can discover and execute.
A restaurant booking form is enhanced with `tool-name` and `tool-description` attributes to make it usable by agents without JavaScript.
The Model Context Tool Inspector extension allows developers to see registered tools, call them manually, and test them with a simple LLM.
By providing a direct API, WebMCP makes agent interactions faster, more reliable, and more efficient by reducing token consumption from parsing HTML.
WebMCP is in an early preview in Chrome Canary, and the team is actively seeking developer feedback on use cases and API design.
The developer extension is available on the Chrome Web Store and is open-source on GitHub for community contributions and feedback.
Adopting standards like WebMCP encourages better semantic markup, making the web more efficient for both agents and crawlers, similar to SEO.
The WebMCP specification is being developed as a draft report in a W3C community group, which is open for public participation and feedback on GitHub.
53:04Alex Nahas
23:17Angie Jones
53:46Chris Heilmann, Daniel Cranney & Raymond Camden
33:51Thomas Steiner
24:12Brad Axen
19:20Jan Curn
25:32Viktoria Semaan
53:54Tarek Ziadé
Jobs that call for the skills explored in this talk.
AVEX Automotive GmbH & Co. KG
Hamburg, Germany
autonomous-teaming
München, Germany
APN Healthcare Solutions
fleetbay GmbH
Koblenz, Germany
Monday Merch
Rotterdam, Netherlands
