AI in Umbraco: What the Codegarden 2026 Session Says About Agent-Ready CMS Work

Phil’s session was not a broad “AI will change everything” talk. It covered how agents can use, build, and integrate with Umbraco outside the traditional backoffice.

The talk covered the pitch phase, backoffice extension skills, content modelling, CMS implementation, custom MCP servers, chained MCPs, remote editor workflows, and collaboration with tools such as Google Drive and Figma.

The useful question has changed from “Can AI generate content?” to “Can an agent help with CMS work without weakening governance?”

What is AI in Umbraco?

AI in Umbraco is the use of AI features, agent skills, and MCP-based tools to support developers, editors, and linked tools working with Umbraco content and code.
The value is not in a prompt that writes a paragraph. The value comes when an agent can use project context, respect permissions, call approved APIs, and complete a CMS task that would otherwise take many manual steps.
Examples include content modelling, backoffice extension work, metadata updates, bulk editorial tasks, media handling, archiving rules, and draft page creation.

Why open source gives Umbraco an AI advantage

AI agents need context. Umbraco has source code, documentation, community packages, examples, and years of public learning material around it.

Phil called Umbraco’s open source nature a strength for AI because models have richer public material to learn from. That does not make AI output perfect, but it gives agents better reference material than platforms where key patterns are hidden from public learning sources.

For enterprise buyers, that helps in a working way. A CMS with strong public documentation and stable APIs is easier to review, test, extend, and govern.

Skills and MCP: why they belong together

Umbraco’s documentation explains the split well: skills provide knowledge, while MCP provides capability. A skill can teach an agent how to build a dashboard or structure a content model. MCP gives the agent tools to interact with a running Umbraco instance.

Agent skills act like guidance files. They can cover back-office extensions, property editors, dashboards, content models, cloud use, Razor implementation, or headless delivery.

MCP servers expose tools an AI agent can call. The official MCP specification describes MCP as an open protocol for linking AI applications with external data and tools.

In Umbraco, that can include reading content types, creating media, querying nodes, calling the Management API, or using custom extension APIs.

Chained MCPs bring CMS work into real workflows

A single MCP server is useful. Chained MCPs go further.

Phil described a pattern where one MCP can use tools from another MCP. An Umbraco MCP could work with a custom extension MCP, a CMS editor MCP, Google Drive, Figma, or another business tool in the same workflow.

For a large content estate, that is where AI starts to look useful. Content may be in Umbraco, brand assets in Drive, designs in Figma, compliance notes in documents, and custom data in another platform. Agent workflows only make sense when those links are designed with access rules, review points, and audit trails.

Editor use cases from the session

Phil’s examples were specific.

Create pages from Drive files

An editor could ask an agent to take room specs and photos from Google Drive, create media items, and add a new room page in Umbraco.

The page can follow the same content model each time, with the editor reviewing before publish.

Archive old content

A scheduled agent could place press or story items older than two years into an archive section on the first day of each month.

That reduces repeat housekeeping without asking developers to change website code for a routine editorial rule.

Find repeated blocks

An agent could review room pages and find blocks that appear again and again, such as galleries or booking summaries.

Editors can then decide what should become reusable content rather than copied content spread through the site.

Update brand terms

A rebrand can create weeks of editorial work.

Phil showed how an agent could update terms across content, media descriptions, and metadata, with human review before publishing.

What enterprise teams should plan before using agents

MCP gives agents the ability to act. That raises governance requirements.

The MCP specification says tool use should include human approval paths, visible tool invocation, and user confirmation for operations. Research on MCP has also flagged security and privacy risks around MCP servers, including creation, operation, and update phases.

Before giving agents write access to Umbraco, review five areas:

1. Read access
   Limit content types, media folders, languages, and environments.

2. Write access
   Separate draft creation, metadata edits, archiving, deletion, and publishing.

3. Approval path
   Route bulk edits, deletion, publishing, and metadata changes through human review.

4. Logs
   Record user, time, source prompt, tool call, affected item, and result.

5. Testing
   Run agent workflows in dev and staging against actual Umbraco structures before production use.

How to prepare an Umbraco platform for AI agents

Start with one repeatable workflow where value is easy to prove and risk is contained.

Good candidates include archiving, metadata checks, asset placement, duplicate block reviews, and structured page creation.

Then turn the editorial rule into written instructions. Name the content types, permissions, approval route, publishing rule, and rollback path. Keep skills under versioning, test them against actual project patterns, and use draft-first publishing where possible.

For enterprise CMS estates, the safest pattern is often: agent prepares, human approves, platform logs.

What Umbraco Looks Like in an Agent-Ready World

As an Umbraco Gold Contributing Partner, we help teams work out where AI can add value without giving up control.

That may mean using an agent to support content work, connect with other tools, or handle a defined CMS task. The setup depends on what needs to happen, who approves it, and what the agent is allowed to change.

The aim is not to add AI for the sake of it. It is to build a workflow that saves time, fits the way the team already works, and keeps permissions, review, and accountability in place.

 Book a Meeting with an Umbraco Partner