Drupal AI Explores Shift From Pages to Agent Capabilities

Websites built with Drupal could increasingly expose controlled actions alongside pages and structured information for AI agents to use, according to Will Huggins, CEO and co-founder of Zoocha. Writing for the Drupal AI Initiative on 4 September 2026, Will describes this as a shift from publishing pages to publishing capabilities. His example contrasts an agent reading product pages with one using authorised functions to search products, check availability, or perform actions on a customer's behalf.

Will points to Drupal's structured-content model and several emerging projects as foundations for that approach. Tool API provides a reusable way to define executable actions with specified inputs and outputs, while MCP Server provides a route for compatible external AI systems to access approved Drupal tools. Outside AI is exploring related questions around agent identity, permissions, reliable execution, tool limits, and visibility into what external agents have done. Tool API and MCP Server remain beta projects, so these components represent active development rather than a finished agent-facing CMS architecture.

The article argues that exposing capabilities should not mean bypassing Drupal's existing governance controls. Will connects the model to authentication, permissions, workflow, Drupal AI Guardrails, and the Context Control Center as ways to limit what agents may access or change and provide appropriate context for their actions. He presents webpages and agent-facing capabilities as complementary interfaces, with Drupal potentially serving as a control layer for both information and permitted actions. Readers can explore the full argument in the Drupal AI Initiative post.