WebMCP lets browser-based AI agents invoke website actions through named tools, rather than identifying fields and typing into pages as a person would. A tool describes an action, such as sending a message or searching products, and the information it requires; the site’s existing code handles execution. Chrome and Edge support the experimental API through origin trials, which require participating sites to add a trial token.

WordPress can expose existing functionality rather than build a separate agent interface. The Abilities API, introduced in WordPress 6.9, defines actions with descriptions, input and output schemas, and permission checks. These abilities can already support REST API access and administrative AI features; registering them as WebMCP tools for visitors would require additional code. Forms can also be marked up with toolname and tooldescription attributes. Elementor has announced planned WebMCP support for Atomic Elements, starting with Atomic Forms.

Adoption remains uncertain: the API is evolving, WebKit opposes the proposal, and Mozilla is neutral. Tools also run with the visitor’s permissions, making prompt injection a significant risk; specification labels for consequential actions are only guidance, not safeguards. A WordPress plugin that abstracts the changing API could reduce maintenance, but browser and agent support will determine whether registered tools are used in practice.

Author of the original text: Miriam Schwab (Elementor)