TYPO3 MCP Server: Edit Content with AI in Workspaces

The Model Context Protocol (MCP) provides a standard way for AI agents and language models to access external systems. For TYPO3, this means editors and developers can retrieve and edit content, page structures, and configurations straight from Cursor, Claude Desktop, or n8n, with no need to build bespoke API interfaces. As our foundation, we use the hauptsacheNet/typo3-mcp-server extension by Marco Pfeiffer.

This fork narrows the original concept to TYPO3 v14 and adds workspace-safe operations , a complete OAuth 2.1/PKCE transport , and a declarative security model you can configure through a YAML file without touching code. The result is an MCP server that runs both remotely via /mcp and locally via stdio in DDEV.

Table of Contents  

Overview  

An MCP server acts as a standardised bridge between a language model and an external system. In a TYPO3 context, this means an AI assistant can create pages, read content, submit workspace drafts, and query records, all through a clearly defined interface that the model understands, with no need to implement its own HTTP clients or authentication logic.

The hn/typo3-mcp-server fork (extension key mcp_server) targets TYPO3 v14 installations and offers a production-ready option for AI-assisted editorial and development workflows. The backend module "User → MCP Server" remains the central hub for token management and client setup, without requiring deep changes to your existing TYPO3 structures.

What has been done  

The starting point is hauptsacheNet/typo3-mcp-server by Marco Pfeiffer , which already provides a working MCP server for TYPO3. The fork narrows the extension entirely to TYPO3 v14: all v12/v13 compatibility paths have been removed, Composer constraints raised to ^14.0, and the whole codebase refactored specifically for v14, including a fix for the defaultPageTSconfig removed in v14 and XLIFF 2.0 ICU support.

On the feature side, we have added OAuth 2.1/PKCE remote transport, a complete backend module with token management, a local stdio mode with DDEV auto-detection, and DDEV stdio fixes. The workspace workflow gained workspace-safe operations , with every tool checking the active workspace context. Further additions include dev-site tools (SiteSettings, ViewHelper reference, CreateLocallang, TCA-MCP-Resources), a live test suite for LLM benchmarks, and a complete CLI mirror that makes every MCP tool runnable even without an AI client.

The security model has been rethought from the ground up: instead of hardcoded permissions, a declarative Configuration/Capabilities.yaml governs the scope of the tools. Removing an entry deactivates all dependent tools, with no code changes required.

Features in detail  

Remote transport uses OAuth 2.1 with PKCE via the /mcp endpoint. For local development and DDEV, a stdio mode with auto-detection is available. Supported clients are Cursor, Claude Desktop, and n8n, with configuration templates provided in the backend module. The CLI mirror lets you call each of the 44 tools directly via vendor/bin/typo3 mcp:<tool> with --json or --plain, which is handy for debugging and scripting without an AI client.

Installation  

composer config repositories.typo3-mcp-server vcs https://github.com/dirnbauer/typo3-mcp-server.git
composer require hn/typo3-mcp-server:dev-main

Frequently Asked Questions  

Conclusion  

An MCP server for TYPO3 is not an end in itself: it makes TYPO3 editable for AI assistants without relying on proprietary integrations. The v14 fork extends the existing extension where it counts: it is workspace-safe, OAuth-secured, usable locally via DDEV, and built around a security model you can adapt without code changes. For teams looking to bring AI agents into editorial or technical TYPO3 workflows, it is a pragmatic starting point.