Austria + TYPO3: Intelligent Government Websites Inspired by the German Model

Germany is investing around 108 million euros in a TYPO3-based Government Site Builder, confirming its choice of technology. Austria can expand this same TYPO3 base in a targeted manner: with the Desiderio design system, MCP, a typed API, a central Abilities Registry, Agent Skills, and an optimised backend – traceable, controlled, and digitally sovereign.

Overview

  • Germany is investing around €108 million in a TYPO3-based Government Site Builder – a strong signal for the choice of technology.
  • Austria is building on Desiderio, adding MCP, a typed API, a central Abilities Registry, Agent Skills, and an optimised backend.
  • Every AI action is checked and logged in TYPO3. Publications are approved by a human in the workspace.
  • The result: unified design, faster onboarding, controllable automation, and fewer system disruptions.
Thoroughly updated in July 2026

This article was originally published on 7 October 2025. It now focuses on the strengths of TYPO3 v14: the Desiderio design system, MCP, a typed API, the central Abilities Registry, reusable Agent Skills, and an optimised backend featuring Record Lists, Agentation, Easy Workspace, and the Visual Editor. AI functions run in a controlled manner within TYPO3. We no longer recommend using a preceding n8n instance for this. Sections on governance, digital sovereignty, and TYPO3 hosting on Vercel have also been added. The existing URL remains unchanged.

Summary: Austria's TYPO3 Ecosystem for AI Agents  

Germany is investing around €108 million in a TYPO3-based Government Site Builder. This is a strong signal for TYPO3 as the foundation for government websites. Austria can leverage this validation and take an efficient, controlled path.

The proven TYPO3 foundation is extended with the Desiderio design system, the MCP server, a typed API, the central Abilities Registry, reusable Agent Skills, and an optimised backend. AI functions work directly within TYPO3, not in a secondary system: every action follows a defined schema, respects permissions, and is logged. Content is only published after human approval in the workspace.

How many websites does Austria's public administration run?

Before unifying, you have to know the landscape. Our source-backed directory counts over 7,900 distinct web addresses belonging to the public administration – compiled from 45,278 official datasets and 27,947 verified allocations. To this day, a central registry is missing. This fragmentation highlights the need for a unified CMS that reduces parallel efforts and can lower operating costs. Read the full analysis, including a searchable list: 7,900+ Public Web Addresses: How Fragmented Austria's Digital Administration Really Is.


Table of Contents  

The Ecosystem

The Six Pillars: Desiderio, MCP, API, Skills, Backend, AI Suite

Backend Optimisation

Record Lists, Agentation, Easy Workspace

And n8n?

A brief explanation – now only as an optional thin client

Governance & Sovereignty

Guidelines, logging, EU law, custom models

Hosting

From sovereign infrastructure to Vercel


The Ecosystem for AI Agents at a Glance  

It is not a single feature that makes the difference, but how they interact: a shared design system, a standardised protocol for AI agents, a typed API, reusable Agent Skills, and a backend that makes both editing and AI more efficient.

The Six Pillars  

Desiderio: The Design System

The foundation for all levels

Desiderio delivers 244 content elements built from 49 typed Fluid components. Using themes based on OKLCH color values, the look and feel can be adapted for any level of administration.

  • Content Blocks: real database columns with backend previews
  • Colour schemes for all federal states via theme toggling
  • Accessible (WCAG) and mobile-first
  • Clean for website, headless, and AI all at once

MCP Server

TYPO3 speaks the agent protocol

Clearly regulated, workspace-compatible access to pages, records, files, and editorial workflows. Around 44 tools – each also available as a CLI command.

  • OAuth 2.1 + PKCE for remote clients
  • Prepare write access inside the workspace
  • Read, translate, import, validate, and publish
  • For Claude, Cursor, development environments, and custom AI agents

API and Abilities Registry

Define once, deliver across multiple interfaces

sg_apicore delivers content as structured data: via REST, automatically generated CRUD endpoints, and generated OpenAPI specifications. The Abilities Registry describes every TYPO3 function once and makes it available via CLI, MCP, REST, and desktop applications.

  • Typed interface contracts with JSON Schema, permission scopes, and risk levels
  • Policy checks and audit logs for every execution
  • Access via token or session without repetitive boilerplates

Agent Skills (Skillflow)

Experiential knowledge as reusable workflows

Skillflow brings Agent Skills to TYPO3: as a SKILL.md folder, importable from Git and searchable via Solr. The skills are bound to workspace stages and become part of the review and approval process.

  • "Build landing page", "Check accessibility", "Prepare translation"
  • Skills as versioned TYPO3 records
  • Review and approval become part of the editorial workflow

Optimised Backend

The agency works faster

Record Lists, Agentation, Easy Workspace, and the Visual Editor make daily work noticeably more efficient – while giving AI agents a controlled interface.

  • Record Lists: Grid, Compact, and Teaser views
  • Agentation: Annotations on elements for Claude, Cursor, or MCP
  • Easy Workspace: One-click publishing, including related records
  • Visual Editor: WYSIWYG editing directly in the actual page view

Netresearch AI Suite

The AI foundation – Open Source

Four extensions from Netresearch DTT GmbH power the AI layer: nr_llm, nr_mcp_agent, nr_vault, and t3_cowriter.

  • Interchangeable models for cloud services, EU providers, and self-hosting
  • Encrypted keys in the Vault
  • AI assistant and CoWriter directly in the backend
Newly highlighted: the Visual Editor

For public authorities, content maintenance is often the biggest hurdle. The new Visual Editor by Matthias Vogel ("kanti", anders und sehr GmbH) addresses exactly this: editors edit text, images, and content elements directly in the actual page view without switching between the website and backend forms. A redesigned element library and drag-and-drop make adding and moving content effortless. Every change is prepared in the workspace and only published after human approval – even if it was suggested by an AI.

Deep-dive technical articles:


Germany Confirms: TYPO3 is the Right Choice for Government  

Germany's €108 Million Investment in TYPO3  

Germany is investing around €108 million in the development, migration, and operation of the new Government Site Builder 11 (GSB 11), which is based on TYPO3 – a powerful validation of this technology choice. The decision by the EU's largest economy confirms that TYPO3 can meet the demands of large public administrations: multisite operations for national structures, long-term maintainability, and a robust open-source ecosystem.

Austria's Targeted Development with TYPO3  

Austria can benefit from this validation and take an efficient, controlled path. The TYPO3 foundation remains, but is targetedly expanded: AI, API, and automation are directly integrated into TYPO3, follow shared rules, and do not require any external third-party systems.

FeatureStandard TYPO3TYPO3 Ecosystem for AI Agents
Content MaintenanceTYPO3 backendVisual Editor directly in the actual page view
Records ModuleStandard listRecord Lists: Grid, Compact, Teaser
APIIndividually builtsg_apicore + Abilities Registry
AI ConnectionExternal tool requiredDirectly via MCP and Agent Skills
AI GovernanceIn the external toolPolicy checks, risk levels, audit logging
ApprovalCustom-built stepEasy Workspace: human approval
Data StorageDependent on external toolsControllable within the TYPO3 infrastructure
Cost StructureOperating multiple systemsOne optimised system

Backend Optimisation: The Agency Works Faster  

The biggest daily leverage is not the website itself, but the backend. Three building blocks make editorial work noticeably faster – while providing AI agents with a controlled interface.

Record Lists (List Types)

Extends the Records module with Grid, Compact, and Teaser views, as well as custom display types. Configuration is done via TSconfig and Fluid templates, entirely without PHP. Status indicators highlight new, modified, and moved records within the workspace.

Agentation

Feedback directly on the element. Users highlight a section on the page and send the selector, comments, and computed CSS properties to Claude Code, Cursor, or an MCP agent. This results in transparent, traceable suggestions instead of undetected changes.

Easy Workspace

One-click publishing. Editors can see all pending changes on a page, select the desired content, and publish it along with any linked records. This binding review step is also where AI drafts are approved by humans.

In addition, the Visual Editor offers WYSIWYG editing directly in the actual page view, complemented by an element library and drag-and-drop. This shortens onboarding times and reduces context switching between the frontend and backend.


And n8n? A Brief Explanation  

The original version of this article put n8n at the centre – a preceding low-code tool used to orchestrate AI workflows. We no longer recommend this setup: content logic, permission checks, and logging should live within TYPO3. In a secondary system, these rules would have to be rebuilt separately and kept continuously in sync.

n8n remains an option – as an optional thin client

n8n can still be useful, but it plays a different role. Since the Abilities Registry describes itself via HTTP, a generic n8n node can control TYPO3 functions. n8n then serves as an optional thin client for cross-system workflows; permissions, content logic, and logging remain inside TYPO3. For AI functions within the CMS, n8n is not required.


Governance, Traceability, and Sovereignty  

For the public sector, the critical question is not "Can AI do this?", but rather "Is it permissible, traceable, and sovereign?". This is precisely where the strength of an architecture directly integrated into TYPO3 lies.

These components are already running in our lab

Desiderio, the MCP server, sg_apicore, the Abilities Registry with policy checking and audit logging, Skillflow, Agentation, the Netresearch AI suite, and the Visual Editor are all up and running in our TYPO3 v14 test environment. This is not just a concept, but a live, testable system. The next step is a pilot project with a public authority in active operation.


Hosting: From Sovereign Infrastructure to Vercel  

For critical core portals, permanently available, sovereign infrastructure remains the right choice – ideally paired with self-hosted or EU-based AI inference. For less critical offerings, other options are available: a complete TYPO3 14.3 instance, including the backend, can now be hosted on Vercel. This is ideal for preview environments, campaign microsites, and predominantly read-only information pages connected to an external, GDPR-compliant database. For highly critical portals with strict latency requirements, potential cold starts remain a dealbreaker.

Case study: TYPO3 on Vercel

How this works technically, which states (database, files, search, cron) need to be externalised, and how you can get your own instance running in just a few steps is explained in our dedicated article: Hosting TYPO3 on Vercel: An Honest Review.


Disadvantages and Challenges  

The legacy AI hosting issue has been resolved

The initial version highlighted an unresolved issue: choosing a GDPR-compliant AI hosting provider. The direct integration now provides a solid technical foundation: nr_llm abstracts model providers and allows selection on a per-task basis – whether cloud services, EU providers, or self-hosted. nr_vault manages the corresponding credentials in an encrypted state.


Thanks to the Open-Source Community  

This ecosystem is built on open-source software created by many individuals and companies. We extend our sincere gratitude to them.

Netresearch DTT GmbH

The entire AI layer runs on their code: nr_llm (LLM foundation), nr_mcp_agent (backend assistant), nr_vault (encrypted secrets), and t3_cowriter (AI CoWriter). Thank you, Netresearch.

Content Types Team

All content elements are Content Blocks. Thanks to Nikita Hovratov and the TYPO3 Content Types Team.

in2code & Georg Ringer

Powermail (in2code) for accessible forms, and News (Georg Ringer) for articles and archives.

dkd & Studio Mitte

Apache Solr for TYPO3 (dkd Internet Service GmbH) and Friendly Captcha (Studio Mitte) for search and bot protection.

anders und sehr & TYPO3 GmbH

The Visual Editor by Matthias Vogel ("kanti", anders und sehr GmbH) and the Blog extension (TYPO3 GmbH).

shadcn, Fluid & Tailwind

shadcn/ui (shadcn), Fluid 5 (Simon Praetorius), and Tailwind (Tailwind Labs) for the design system.


Conclusion: Learn from Germany, Think Ahead in a Controlled Manner  

Germany's investment of around €108 million confirms TYPO3 as a robust, enterprise-grade foundation for government websites. Austria can leverage this signal and take the next step: running AI, APIs, and automation directly, securely, and sovereignly within TYPO3.

The building blocks: the Desiderio design system, the MCP server, a typed API, the central Abilities Registry, reusable Agent Skills, and an optimised backend featuring Record Lists, Agentation, Easy Workspace, and the Visual Editor. Data flows remain fully controllable, credentials lie encrypted in the Vault, and models run either in the EU or on-premises. This allows the proven TYPO3 foundation to be expanded in a targeted, traceable, and secure manner.


Resources and Further Reading  

For strategic discussions: office@webconsulting.at

Let's talk about your project

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.