Bringing Drupal AI into your DNA: Creating your own AI Agent MCP workflows without Code!

In the final session of the Bringing Drupal AI into Your DNA series, the focus shifted from theory to accessibility. After earlier sessions explored the basics of Drupal AI, search, and developer-led agent creation, this last training looked at how AI agents can become easier for more people to build and use.

The session began with a reminder of the bigger goal behind the series. The Drupal AI initiative was launched to help people not only understand AI, but also experiment with it, contribute to it and apply it to their own work. At the centre of this effort is the AI module, which acts as an abstraction layer and makes it possible to connect Drupal to a wide range of AI providers.

This final session focused on agents again, but from a more practical angle. The main question was no longer only what AI agents are, but how people can actually create them more easily.

A large part of the session was dedicated to a new visual workflow interface for Drupal AI agents. Instead of relying only on code and configuration spread across different parts of Drupal, users can now build agents in a more visual and intuitive way. In the demonstration, a simple agent was created to check the fields available in a content type. The process was straightforward: create the agent, assign it a task, add the tool it needs, connect it to an assistant, and then test it through a chatbot placed directly on the site.

What stood out was how clearly this showed the full chain. The chatbot is what the user sees. Behind it sits an assistant, which manages the conversation and context. The assistant then calls the agent, and the agent uses the selected tools to carry out the task. What used to feel quite technical suddenly became much easier to follow.

Another example brought this closer to a real editorial use case. An AI agent was created to evaluate image alt text on media items. In the demonstration, an image had a very poor alt text value simply set as “test”. The agent reviewed the image, checked the existing alt text, generated an explanation, assigned a quality score, and then updated the media item by saving the result as a draft. This is useful because it does not remove human control. Instead, it helps editors quickly identify weak alt text and improve accessibility across large volumes of existing content.

This example made the value of agents much more tangible. Rather than producing generic text, the AI was supporting a concrete quality review task that many content teams struggle to carry out consistently at scale.

The session also highlighted an important point: if Drupal is going to use AI more widely, it also needs good governance. One of the advantages of doing this work in Drupal is that permissions, roles and access checks can be built into the system. Agents can be limited to specific tasks, tools can check permissions before acting, and site owners can have much more visibility over what is happening.

This matters because AI agents are powerful. They can retrieve information, edit content and trigger actions. That is why the discussion also touched on logging, auditability and guardrails. These are the checks that help ensure AI behaves within clear boundaries, whether that means respecting permissions, avoiding harmful content or following organisational rules.

The session ended with a look at what is coming next. Drupal is moving towards tighter integration between agents, workflows and external systems. There was also a glimpse of how AI might eventually help users build other agents, reducing the technical barrier even further.

The main message of the session was encouraging. AI agents in Drupal are no longer only for developers. With the right interface, clear permissions and practical use cases, they can become a useful tool for a much wider group of people.

This final session closed the series on a strong note. It showed that Drupal is not only experimenting with AI, but also thinking carefully about how to make it usable, governable and relevant to everyday work.

If you are interested in knowing more, you can rewatch the session here.