Model Context Protocol MCP for technical writers
It’s November 2025 and I just came back from this year’s tcworld conference. Pretty much a year ago in November 2024 Anthropic introduced an open-source standard, the Model Context Protocol MCP. It allows connecting apps and servers to generative AI such as ChatGPT, Claude or Gemini.
Technology moves fast these days; generative AI moves lightning fast. And a year was enough for developers in all domains to develop thousands of MCP applications. When I first heard about MCP servers earlier this year, I immediately had to try some of Anthropic’s reference implementations. Seeing the LLM doing actual work on my laptop was eye opening.
Half a year later at tcworld, I had the opportunity to share that experience and present some MCP servers that might be helpful for technical writers. While the use cases that I picked are not going to cut anyone’s workload in half, they illustrate what generative AI is capable of if combined with MCP servers. After the talk, I was approached multiple times about MCP and had some interesting discussions about opportunities and risks of this new technology. That’s why I thought it might be a good idea to share some use cases here on our blog. If you haven’t heard about MCP servers, the following videos might give you an idea about what’s possible.
Manipulating folders
The following video illustrates how to set up a folder structure for a docs-as-code project. The folder structure reflects the product families with product variants. I’m using Claude Desktop with Anthropic’s filesystem MCP server. When using an MCP server the first time, you will be asked to grant access to the MCP server and its tools. I had allowed it previously, so in the video we don’t see that.
Reading PDFs
But MCP servers can do way more than creating folders. In this video I am using the pdf-reader-mcp in combination with the filesystem MCP server to work with PDFs. You will see that Claude Desktop is deciding which MCP server to use for which task.
Creating DITA content
MCP servers can be used with different LLM providers and different clients. Now we are going to use the same MCP servers as before but with ChatGPT within Oxygen XML Editor and the AI Positron Assistant. We will see that ChatGPT requires slightly different prompting than Claude. And we will also see that your token limit can never be too high.
Manipulating 3D models
But MCP servers can also automate software. In the following video we will see how blender-mcp manipulates a 3D object. The MCP server requires a Blender plugin to connect. Also, having a high-quality model with good naming of its parts helps the LLM work with the model. In the video, you will see an industrial robot from KUKA available on blendswap.
I hope you got an impression of what you can do with MCP servers. And with thousands of MCP servers available, that was only a fraction of all the possibilities. Apart from the tools that we saw in the videos, MCP servers can also provide additional resources and standardized prompts. And while the non-deterministic nature still requires a human in the loop, generative AI and large language models are getting more and more ready for production.