AI in Umbraco in 2026: choices, governance, and day-to-day use

What Umbraco has put on the table

Umbraco’s AI work spans two areas.

One area involves AI working with Umbraco through tools that enable an AI assistant to interact with the back office via supported interfaces. The Developer MCP Server is part of that direction, with documentation for Umbraco CMS 16 showing how the MCP server is configured and used, plus a safety warning about production usage.

The other area covers AI inside Umbraco, where AI support appears in editorial screens through packages and configuration. Umbraco’s own AI page describes the direction as optional, integration-friendly, and centred on developer choice.

Those two sides can make a lot of difference for enterprise buyers because they map to two different needs:

• Editors can have assistance with content and media work.

• Platform and engineering teams get a predictable integration point, auditability, and a way to keep tools aligned with the Umbraco version in use.
  

Editorial and media production: where AI actually helps

A general “AI feature” is the last thing your editorial team wants. The field-level approach of Umbraco can cut short a lot of repeated tasks like creating image text, page summaries, meta descriptions, content review, and consistency checks.

With AI in Umbraco, the Content screens the editors look at stay the same, and with that, a prompt is available for a specific field, and they can preview the output and decide what to choose

And thats how most companies want to govern their content. While Human reviewing everything, the AI output comesce one more input to review. 

A practical media example shown during the Umbraco Winter keynote was alt text which is a great deal for for accessibility and search snippets. It also matters because alt text is a place where teams often accept “good enough” due to time. AI help here tends to lift baseline quality, as long as the editor keeps final control.

Meta descriptions are another example. Many teams either skip them or write them late. AI assistance inside the field can speed that part up, again with editorial review.

Provider selection and procurement alignment

Legal, security, and procurement teams usually define which AI model providers can be used and under what terms. These providers are the services that run large language models, such as OpenAI, Google Gemini, Anthropic, Amazon Bedrock, and Microsoft Azure or Foundry.

Umbraco supports this through provider packages installed from the Umbraco Marketplace. Each package connects Umbraco to a specific AI provider using that provider’s own API and credentials. These packages sit alongside the CMS and can be enabled per project.

More than one provider package can be installed within the same Umbraco project. Tasks inside the CMS can then be assigned to the provider that fits internal rules and the type of work being done. Editorial assistance can use one provider, while development-related tasks can use another.

This approach avoids rework when provider rules change. Content stays where it is. Editorial structures stay intact. Teams adjust which provider a task uses, rather than rebuilding pages or changing how content is stored.

During the Winter Keynote, Umbraco explained this provider pattern and confirmed support for multiple providers, with OpenAI used as the demonstration example. The broader list of providers reflects the intended direction rather than a single fixed setup.

For business owners, this leads to three outcomes:

• Provider selection follows approved contracts and internal rules

• Different tasks can use different AI models, where permitted

• AI usage costs follow the selected provider instead of being bundled into the CMS license

Profiles and context: controlling behaviour per task

In the case of Enterprise website content work, they have very different and unique tasks. Image text and SEO descriptions are not the same as policy pages, public notices, or product content.

Umbraco’s keynote demo described two configuration ideas that map well to enterprise realities:

• Profiles for task-specific behaviour (example: text review vs image analysis).

• Context for persona-based behaviour (example: a marketing persona with a defined tone and writing rules).

This approach reflects how governance is applied in real environments. Roles and tasks usually require different handling rather than one shared setting for all content.

A useful way to think about it:

• Profiles match the task.

• Context matches the user and output expectations.

That structure gives organisations room to apply AI support in selective areas while keeping other areas tightly governed.

Logging and review: audit expectations

When AI is used in Umbraco, it produces text, image descriptions, summaries, and other content that may appear on public pages. Once that content is published, it becomes part of official communication and must follow the same review standards as any other material.

During the Winter Keynote, Umbraco showed analytics and logs within the AI area of the backoffice. These records capture when AI is used, which task it supports, and how often it is applied. They exist to support review and follow-up, rather than to automate decisions.

Editors continue to review AI-assisted content as part of their normal publishing work. Suggestions appear in the same places where content is written or edited, and decisions are made before anything is published. The activity records then remain available if questions come up later about how a page or media item was prepared. This keeps review, accountability, and follow-up aligned with existing content ownership.

Developer tooling: MCP and “skills”

For engineering teams, AI value appears when tools reduce friction in day-to-day implementation and maintenance.

Umbraco’s MCP work is a concrete piece here. Umbraco has published an MCP server article describing it as a way for AI clients to collaborate with the CMS, and Umbraco documentation includes guidance for the Developer MCP Server.

The documentation includes a strong production warning. That detail matters because it signals a security mindset: AI integrations require careful environment boundaries.

The keynote also introduced “skills” as a way to help coding agents produce implementation work aligned to Umbraco patterns, starting with backoffice extension work. This is a developer-side governance problem: keeping generated code aligned with the Umbraco version and expected patterns.

Winding Up

AI works well in an Umbraco setup when it is introduced in a way that respects how the platform is already used. Content and media follow the same structure people are familiar with, and day-to-day work continues without new patterns being forced into place. AI support appears only in selected areas, where it helps with specific tasks, and it stays out of parts of the platform where it adds no value.

This approach suits Umbraco installations that are expected to run for years. These platforms usually grow across multiple sites, regions, and teams, and decisions made today tend to stay in place for a long time. By keeping AI optional and configurable, Umbraco allows organisations to add support gradually while keeping ownership, responsibility, and platform behaviour predictable.

At Phases, this is the way we approach Umbraco work. As an Umbraco Contributing Gold Partner, we usually begin by understanding how the platform is already used, then look at where AI support could help without changing established ways of working. Editors, engineers, and platform owners are involved early so that decisions reflect real usage rather than assumptions.
If AI is part of your thinking around Umbraco, it can help to speak with a partner who works with these platforms every day. Not to introduce something new for the sake of it, but to look at what already exists and decide where support makes sense.

Get Started