Previous
About the author
- Albert Skibinski is a freelance full-stack developer en co-founder at Jafix.
- I write about web development, long bike rides and food!
MCP AI Drupal
Using MCP to Manage Content with Ai Agents
Albert Skibinski •
In my previous post about Model Context Protocol (MCP) with Drupal, I explored the theoretical possibilities of connecting AI agents to Drupal sites. Now, I'm excited to share the follow-up: we've built a MCP server in Drupal that lets Ai agents directly manage tutorial content.
Here is a demo how I connected Claude.ai (webbased) to Jafix MCP. Note that you need a payed subscription to add this.
In this case, we let Claude create a demo tutorial by itself, but you can imagine that you could paste in raw text and let ai figure out the appropriate tools to create nicely structured content (paragraphs, various fields, etc).
A new way to create and manage content
Our MCP server exposes a comprehensive set of tools that allow AI agents to:
- Manage tutorial content: Create, read, update, and delete repair tutorials
- Handle complex hierarchies: Organize tutorials by brands, models, and issue types
- Process media uploads: Handle images and videos for tutorial steps
- Support multiple languages: Create translations of existing tutorials
- Manage taxonomies: Work with brands, models, issues, and object categories
How does the Ai "know" howto use the tools? We use a "describe" tool which returns extensive information on fields and how content should be structured, including examples.
The result? An ai agent that supports MCP can now act as a content management assistant, helping create and organize repair tutorials through natural conversation.
This allows users to use their favorite Ai tool to manage and generate content on Jafix. They can still use the webbased interface on Jafix.com of course, but if they prefer they can manage it by talking to an Ai.
MCP module
The Drupal MCP contrib module provides a plugin system. You create plugins that implement the Mcp interface, and they're automatically discovered and exposed to AI agents.
The contrib module currently provides two different transport layers. See the docs.
- TypeScript binary (
mcp-server-drupal) handling stdio transport (for local) - HTTP API endpoint for JSON-RPC communication
It doesn't support SSE, but that has been deprecated anyway in favor of Streamable HTTP which is implemented partly. However, I had some difficulties getting Claude.ai to talk to our MCP bumping into some issues:
- Tool names were too long because of hashing (issue)
- It seems claude.ai uses an older transport protocol version (2024-11-05 versus 2025-03-26) which required some patching.
Note though that developments are moving fast in this area, so this will probably change soon.