What if your Drupal site could talk to your project management, ticketing, CRM, email, calendar, voice assistant, and smart home — through one AI-orchestrated workflow? We built exactly that. This session shows how a small agency wired Drupal into a full-scale agentic operations platform using MCP, CLI tools, and a Rust core engine — and runs it in production every day.

Prerequisite

Comfortable with Drupal development (module installation, configuration, basic architecture). Familiarity with APIs and command-line tools is helpful. No AI/ML background required — we'll explain the agentic patterns from the ground up.

Target Audience

Developers, technical leads, and agency CTOs who want to see what a production agentic platform looks like when built around Drupal's MCP ecosystem. Ideal for anyone evaluating how to connect Drupal with external systems through AI orchestration.

Outline

Our agency built Houston - an agentic AI operations platform in Rust that puts Drupal at the center of a 10+ system constellation. Through MCP servers, CLI tools, and custom integrations, an AI assistant can create Drupal content, book time in OpenProject, triage Zammad tickets, generate CRM invoices, search emails, check calendars, query public transport, control smart home, and read vehicle status - all from a single message in Mattermost.

Houston also closes the loop from ticket to Drupal code: an OpenProject ticket for a new content type, field, view, custom module, or theme tweak is read as spec, a feature branch is created in Git, the Drupal code is written against the project's codebase, and a merge request is opened. A confidence score decides whether Houston asks for review or flags uncertainty.

The core includes a scheduler running 50+ tasks, vector memory for prompt injection, a learning engine, and a real-time voice pipeline (Whisper → AI → Piper) with German and English voices.

We'll cover: mcp_server and mcp_tools in production - what works and what doesn't. The integration stack beyond Drupal. The autonomous ticket-to-Drupal-code pipeline with confidence gating. GDPR-compliant model routing with EU models (Mistral) and local inference (Gemma4, Phi-4) on Apple Silicon. Interfaces - Mattermost, web, voice, CLI, REST, mobile GPS. The open/commercial boundary: building blocks are open, orchestration is ours.

It all runs daily. Expect a live demo.

Learning Objectives

1. How to expose a Drupal site as an MCP server and what AI agents can do with it today
2. A production architecture for orchestrating Drupal alongside 10+ systems using MCP servers, CLI tools, and a Rust-based core engine
3. How voice interfaces, scheduled automation, and semantic memory transform an AI assistant from a chatbot into an operations platform
4. Strategies for GDPR-compliant AI deployment with EU-hosted and local models
5. Where the line is between community contribution and commercial product — and why that's healthy for the Drupal ecosystem

Experience level
Anyone