TYPO3 MCP Server: Setup, Daily Use and Practical Examples

Translating a page, adding missing meta descriptions, correcting a changed phone number on forty subpages: tasks like these take many clicks in the TYPO3 backend, even though they can be described in a single sentence. This is exactly where the TYPO3 MCP server comes in. The extension connects TYPO3 via the Model Context Protocol to AI agents such as Claude, ChatGPT, Le Chat or Gemini. Editors describe their task in a chat, the agent does the work in the CMS – and every change lands in a workspace first instead of going straight to the website.

In this article we show how the extension is set up, what daily work with it looks like and which tasks it is particularly worthwhile for in practice.

What the MCP server does in TYPO3

The Model Context Protocol (MCP) is an open standard that AI applications use to work with external systems. An MCP server provides the agent with clearly described tools. In TYPO3 these include functions to read the page tree, retrieve pages with their content elements, search records, understand table structures, write records and upload files.

The agent does not work around TYPO3 but with its built-in tools. It follows the TCA, so it knows the same fields, content types and relations as the backend. This also applies to custom content elements and extensions, provided they are configured properly and are workspace-capable. Whatever can be edited in the backend can usually be edited by the agent as well – with exactly the permissions of the signed-in backend user.

The most important security principle: all content changes are automatically staged in a TYPO3 workspace. The agent cannot publish anything. Only once a responsible person has reviewed and approved the changes in the preview do they become visible to visitors.

Requirements

Before installing the extension, check the starting point:

• TYPO3 13.4 LTS or 14 – older versions are not supported.
• PHP 8.2 or higher.
• The core Workspaces extension. It is installed automatically as a dependency.
• A Composer-based installation for managing extensions.
• An MCP-capable client, such as Claude, ChatGPT, Le Chat, Gemini or a developer tool like Cursor.

If a website still runs on TYPO3 12 or older, a TYPO3 update comes before the integration. Both can easily be combined in one project.

Installation and configuration

Install the extension

The extension is installed via Composer like any other package: composer require hn/typo3-mcp-server. Then run the database analysis and clear the caches as usual. There are no further mandatory steps – the default settings work without adjustment.

Check the extension configuration

The settings are found under Admin Tools → Settings → Extension Configuration → mcp_server. Three options are relevant:

• Maximum file size (default: 500 MiB) for files the agent loads from a URL or receives via an upload link.
• Additional read-only tables (default: sys_file). This lets the agent search the file list without changing it.
• Additional standalone tables (default: sys_file_metadata). This makes file titles and alternative texts editable, for example.

For most websites these values can stay unchanged. If you want to expose further extension tables as read-only, add them here.

Prepare workspace and permissions

The MCP server uses the first workspace the signed-in user is allowed to write to. If none exists and the user may create workspaces, the extension automatically creates its own workspace named “MCP Workspace for” followed by the user name.

In practice we recommend creating the workspace deliberately and giving it a meaningful name, such as “AI Drafts”. That way the whole team can see which changes come from the agent. A suitable permission concept for AI agents is just as important: the agent inherits the permissions of the connected user. An editor account with restricted page and table permissions is therefore the safer choice compared to an administrator account.

Connecting AI agents to TYPO3

Option 1: Connection via OAuth (recommended)

For daily work, connecting via OAuth is the recommended way. It also works for websites on a remote server and relies on the normal backend login:

1. In the TYPO3 backend, open the user menu at the top right and select MCP Server.
2. Copy the displayed server URL, optionally also the suggested integration name.
3. In the AI client, create a new MCP integration or connector and enter the URL.
4. On first access the TYPO3 login opens. After logging in, the agent is connected with this backend user.

From then on the agent only sees the pages and records the user has access to and writes exclusively to that user’s workspace.

Option 2: Local connection via the command line

For development environments there is also a local connection. The AI client starts the MCP server directly via the TYPO3 console command vendor/bin/typo3 mcp:server. This option runs with administrator permissions by default and therefore belongs only on local development or test systems, not on a production website.

How editors work with the agent

Day to day, working with the MCP server follows a simple process:

1. Describe the task: the editor describes in the chat what should happen – just as she would explain it to a colleague.
2. The agent gets an overview: it reads the page tree, retrieves the affected pages and checks which fields and content types are available.
3. Changes in the workspace: the agent creates, changes or translates content. Everything is staged as a draft.
4. Review in the preview: in the Workspaces module or via a preview link, the drafts are checked in the real layout – by colleagues without backend access, too, if desired.
5. Publish: if everything fits, the changes are approved with one click. Anything that does not fit is discarded or improved in the chat.

A few habits noticeably improve the results. Full page URLs help the agent find the right page immediately. A clearly defined scope – “only the pages under /services” – prevents unwanted changes elsewhere. And for larger projects it pays to proceed step by step: first one page, then the review, then the rest.

Practical examples

Translating pages

“Translate the page /careers into English.” The agent reads the content, translates it and creates the language versions in TYPO3’s language structure – as real translations of the source elements, not as loose copies. The translation function is still marked as experimental in the extension. A professional review before approval is therefore a must, especially for technical terms and legal statements.

Maintaining SEO metadata

“Which pages have no meta description? Add suitable descriptions.” The agent searches the page tree, writes descriptions based on the actual page content and enters them in the page properties. The editorial team then reviews them collectively in the workspace instead of opening each page individually.

Bulk changes

“Replace the old hotline everywhere with +49 221 29 83 12 80.” Changed phone numbers, renamed products or outdated links are typical tasks where it is easy to miss a spot in the backend. The agent finds all occurrences and changes them in one pass. The workspace overview then shows exactly which text stood where before and after.

Turning documents into pages

“Turn this flyer into a product page.” Word documents, PDFs or briefings become structured pages with suitable content elements. The agent chooses from the content types the website actually offers – including custom elements.

Content audits

“List all images without alternative text.” Not every task has to be a change. Agents are very well suited for inventories: missing alternative texts, duplicate content, outdated years or pages that still contain the old company name. The result is a prioritised list as the basis for the next editorial round – also a good contribution to accessibility.

Embedding images, videos and files

“Put this YouTube video on the home page.” The MCP server can load files from a public URL, create YouTube and Vimeo links as online media, store text formats such as SVG or CSV directly and receive files from your own computer via a single-use upload link. Existing files are never overwritten or deleted.

Migrating content during a relaunch

“Transfer the news from the old system into the new news section.” During relaunches and migrations the agent can take over content from legacy systems and slot it into the new page structure. This does not replace proper migration planning, but it saves a lot of manual transfer work.

From our own experience: building new pages

We use the MCP server ourselves. We created the landing page on TYPO3 and AI agents via the agent – including the hero section, tiles, FAQ and call-to-action elements. First in the development environment, then as a draft on the live website, where it was reviewed and approved like any other change.

Limitations you should know

As helpful as the MCP server is, a few points should be clear before productive use:

• Files are not versioned. Uploaded files are stored in the file system immediately. They only become visible once approved content uses them.
• No workspace selection. The extension uses the user’s first writable workspace. If you run several workspaces, take this into account when assigning permissions.
• Translations are experimental. They work but should be reviewed carefully before publishing.
• Data protection depends on the agent. The content stays in your own TYPO3. Which data is transferred to the AI provider depends on the client you choose. Data processing agreements, storage location and use for training should therefore be clarified in advance.

Conclusion

The TYPO3 MCP server turns the AI agent into a real tool for website maintenance. Setup is done quickly with one Composer command and an OAuth connection. The real difference, however, comes from the guardrails: workspaces, user permissions and preview ensure the agent takes work off your hands without taking control away.

Would you like to know whether your TYPO3 website is ready for AI agents? We check version, workspaces and permission concept and set up the integration for your team. Find out more on our page about TYPO3 with AI agents – or request a free website analysis directly.