ECA Guide Supports AI-Assisted Workflow Building in Live Demo

During a Drupal AI Learners Club session on 31 July 2026, an AI agent used machine-readable documentation for Drupal's ECA module to build and validate workflow models. ECA maintainer Jürgen Haas presented the ECA Guide as a documentation system for both people and automated tools. The recording extends earlier coverage of Haas's documentation work with live model-building examples.

The demonstration showed how an agent could retrieve ECA information without inferring configuration from PHP or scraping navigation-heavy HTML. Haas said the 1,364-page guide generates most plugin reference material from the codebase, while handwritten content adds concepts, caveats, and common mistakes. His comparison put one plugin question at more than 600,000 tokens when working from PHP, about 154,000 from the guide's HTML, and 161 from the underlying Markdown. Haas noted that some agents may strip HTML before passing content to a language model, reducing the effective model-input cost. The guide exposes machine-readable material through Markdown pages, an llms.txt index, an agent skill, and a Model Context Protocol server.

In the first live example, an OpenCode agent searched the MCP server for suitable plugins, retrieved their configuration schemas, assembled an ECA model as YAML, and sent it back for validation. After the validator returned warnings, the agent revised the model and submitted it again, reaching zero errors and zero warnings while also handling a case in which previously published content was unpublished and later republished. Haas later demonstrated an in-progress social-promotion recipe comprising eight main ECA models and a temporary ninth model used for testing, with additional data stored separately and external posting delegated to Activepieces. The documentation repository includes the MCP server and MkDocs configuration used for the guide.