Most people think of the Model Context Protocol (MCP) as something only an AI harness like Claude or ChatGPT uses. It's true that attaching an MCP-enabled resource to your AI tool of choice is easy. But MCP can do much more than that. When you combine Drupal, Maestro and MCP, MCP-enabled websites become a live source of data and actions for a real workflow engine. That engine is deterministic, it puts a human in the loop, and it runs inside your Drupal site.

That combination is the point of this post. AI is very good at producing content and reasoning over data. It is not good at guaranteeing that the right person approves the right thing in the right order every time. That is what Maestro does.

Why Maestro for Human-in-the-Loop AI

AI agents are probabilistic by nature. Ask the same question twice and you may get two different answers. That's fine for drafting a paragraph. It isn't fine for publishing content, approving spend, onboarding a customer or anything else your organization is accountable for.

Maestro is Drupal's business process workflow engine, and it brings the structure these processes need:

• Deterministic flows: every step, branch and condition is designed up front, so the process runs the same way every time.
• Human-in-the-loop tasks: people review, edit, approve or reject at exactly the points you choose, before anything moves forward.
• Assignment and permissions: tasks go to specific users or roles, and workflows run only for people allowed to start them.
• Visibility: you always know where a process is, who it's waiting on and what has already happened.

Let AI do the heavy lifting inside a step, and let Maestro decide what happens next. That's how you get the speed of AI without giving up control.

Maestro Workflows as Tools

Maestro has been "tool" capable for quite some time. In Drupal, tools declared through the Tool API are available to any other tool-capable module. Each tool has a data contract between the caller and the callee that defines its inputs, outputs and data formats. Maestro can expose any workflow flagged as public as a tool. Any tool-capable mechanism Drupal supports can then launch that Maestro workflow without bridging code, and without knowing anything about how Maestro works.

You probably don't know how your car's ignition system works, yet you turn the key or push the button every day. The ignition is well defined, self-contained, well tested and does one job. A Maestro workflow is no different when it's properly designed, secured with permissions and exposed as a tool for others to use.

Maestro Workflows Exposed to AI Agents with Drupal MCP

Once something is exposed as a tool in Drupal, Drupal's MCP implementation can make it available outside Drupal. External AI agents and other MCP-capable applications can then use your Maestro workflows. A Maestro approval flow that was once self-contained becomes a tool with a defined data contract that any AI agent or application can call.

The agent can start the process, but Maestro still runs it. Your approvals, assignments and business rules apply no matter who or what started the workflow. This opens up a whole new range of possibilities for Maestro and for Drupal in general.

Maestro Calling External MCP Tools

It works in the other direction too: Maestro can call tools. Tools can come from ECA, from FlowDrop, or from a remote MCP server connected through Drupal's MCP Client. Wherever they come from, they show up in Drupal's tool ecosystem, ready for Maestro and any other tool-capable module (ECA, FlowDrop, AI and more).

So a Maestro workflow can pull data from an external MCP server, hand it to AI to generate content, and then send the result to a person for review before anything is published. That gives you a repeatable, auditable process.

Video: Drupal, Maestro and MCP in Action

I put together a video that uses an external MCP-enabled website as the data source for AI-generated content inside Drupal. Dries Buytaert kindly volunteered his own blog as the proving ground for this real-world scenario.

In the video you'll see:

• How Dries' MCP server provides tools that Maestro can use
• How ECA orchestrates the Drupal events
• How Maestro acts as the overall human-in-the-loop orchestrator, so a person stays in control of what AI produces

The Takeaway

MCP isn't just for chat windows. With Drupal and Maestro, MCP connects AI agents, external websites and Drupal into one controlled workflow. AI generates, external systems supply data, and Maestro makes sure every step happens in the right order, with the right people signing off. If you're putting AI into processes your organization depends on, build them on a deterministic workflow engine. In Drupal, that's Maestro.