Umbraco MCP Quickstart: Setup in 5 Minutes (with Real Examples)

  • Created by ProWorks
  • Duration: 8:07
  • ProWorks Corporation
  • Watch on YouTube

This video shows how to set up the Umbraco MCP server locally with the test site, API user, and VS Code Copilot configuration. It also demonstrates using AI through MCP to review and update image alt text, SEO metadata, and new content structures.

Setup starts with cloning the Umbraco MCP GitHub repository and building its included test site from the command line. The first run fails until you add an appsettings.Local.json file. After that, Jason installs Umbraco, creates a user group and an API user with client credentials, and adds an mcp.json file so VS Code and GitHub Copilot can reach the Umbraco Management API.

Once connected, he works through real prompts: auditing images for missing alt text, refining the prompt until it returns only the images that need it, and writing the new alt text back to the media items. He then has Copilot add missing meta descriptions and build a new news section from the existing blog document types, asking it for a plan before letting it make the bigger change.

Key takeaways

  • The video shows how to get a local Umbraco MCP environment running by cloning the repo, building the test site, and fixing the initial configuration issue with a local settings file.
  • Viewers learn that MCP connects through the Umbraco Management API and requires a properly permissioned user group, API user, and saved client credentials.
  • The walkthrough demonstrates how to connect VS Code and GitHub Copilot to the MCP server using a project-level mcp.json configuration file.
  • It illustrates that prompt quality affects results, especially when identifying only the images missing alt text instead of updating every image.
  • The example SEO workflow shows AI retrieving documents, evaluating gaps, and writing meta descriptions while leaving existing content intact.
  • For larger structural changes, the video recommends asking AI to produce a plan and questions first before creating new content areas based on existing document types.

Key moments

  1. 0:00 Introduction and what Umbraco MCP does
  2. 0:39 Clone the repo and run the test site
  3. 1:48 Install Umbraco and review the sample content
  4. 2:37 Create the user group and API credentials
  5. 3:16 Configure VS Code, Copilot, and mcp.json
  6. 4:48 Audit and update missing image alt text
  7. 6:06 Review and generate SEO metadata
  8. 7:02 Create a new news area from existing doc types

Transcript

Show transcriptHide transcript

[0:00] I'm Jason from ProWorks. We're an Umbraco Platinum Partner, and today I'm really excited to show you the Umbraco MCP server: how you can get it set up to have a play around with it, and give you some ideas of how maybe you can use it in your projects. Let's go. If you're interested in learning more about Umbraco MCP, Philip at HQ has a great blog, and it's a good place to start. He introduces how the Umbraco MCP server works, shares where you can read more about it, and presents several real-world use cases that may inspire you. The Umbraco MCP server allows Copilot and other AI tools to access the Umbraco Management API and make changes to Umbraco. Let's set up the Umbraco MCP server. In order to set up an Umbraco MCP server locally, we first need to go to its GitHub repository.

[0:50] Now, copy the repository URL so we can clone it locally. After it's finished cloning, you'll notice that there's an MCP server and a test site folder. We're going to use the included test site in order to test out the MCP server. You can use the command line to build the test server. Then, inside the MCP test site directory, we'll run it using dotnet on the command line. Unfortunately, as you can see here, we get an error when we try to run it. We'll need to create an appsettings.Local.json file. I've opened the MCP test site folder in VS Code, and I'm making a copy of appsettings.Development.json and renaming it to appsettings.Local.json. You may need to enable HTTPS for development. Then we can execute dotnet run, get the website running, and grab the URL that we can test with.

[1:48] Let's take that URL and install Umbraco. I don't want to assume any prior knowledge of installing Umbraco, so I'll run through the installation steps now, real quick. Then, after a quick login, we can start looking at the example content. The test site that's included with the Umbraco MCP server repository is based off of Paul Seal's starter kit. It has a nice set of blogs and example pages and media that we can use to test with. One thing I've noticed is that there aren't SEO meta settings on many of the pages. I'll try to fix this with AI later in the video. In addition, I've noticed that some of the media items have alt text and some do not. And here's a quick look at the front end, so you know what it should look like.

[2:37] Umbraco MCP connects to Umbraco through the Management API. First, we'll create a new user group to allow it to access the content and media areas of the website. Go ahead and allow access to all languages, content and media. Then just set all the permissions to true. Now we can create an API user using that user group. Allow access to documents and media. And this is the key part: we need to create client credentials so that the MCP server can authenticate and access Umbraco. Save this secret key so you can use it later in the process. Okay, we're ready to configure VS Code to connect to the MCP server.

[3:22] Let's run VS Code as an administrator. Find the GitHub Copilot extension and install it. This area on the right side is where Copilot will run, and this box is where the prompts will go. Next, we'll configure VS Code to use our MCP server. Create a new .vscode folder in the MCP test site. Then add an mcp.json file. The Umbraco MCP GitHub has an example configuration file that we can use, so we'll copy and paste it now. Then set the client ID and client secret configuration settings to the values that you used for the API user earlier. Switch Copilot to agent mode in order to run MCP tools. Then start the Umbraco MCP server running. You may need to sign in to GitHub to use Copilot. If you get an error about having too many tools, you can go to the tools selection and remove tools you don't need. Now we can ask it for the homepage GUID, just to test, and it will ask us for permission to access the API. And we can approve that, and there's the homepage GUID. So the test worked, and we have access.

[4:48] Now we get to the fun part. Let's see what MCP can do. First, I'd like to address the issue with the alt text on images. So, we'll go through some prompts to try to identify the images with problems, and see if AI can fix it. The results from this prompt weren't great. So, let's keep going and see if we can get more detail. This time, we'll get a report of all the images, whether they have alt text set, and what folder they're in. Well, that's better. Maybe it can write some alt text for us, so let's try that next. This next prompt worked, but I wasn't clear enough, and it added alt text for all images, not just the ones without alt text. And it looks like with this prompt, I got just the images that don't have alt text, with its recommendations. Now, let's see if AI can update Umbraco and set the alt text for us. This is promising. It looks like it thinks it did it correctly. So, let's go to the backoffice and see how it did. Yeah, looks like it did it. Nice job, AI.

[6:06] Okay, now let's turn to the content and see if AI can tell us what content pages have good SEO properties, and maybe help us set them as well. This looks like good information, and I appreciate the summary at the bottom. So now let's try to get AI to set up some valid SEO meta properties for our pages. It's calling the Umbraco API to get all the documents, then updating all the documents, hopefully keeping everything there and simply adding the SEO information. Looks like it's done, so let's see how it did. The Content tab still has everything, and SEO has a meta description. Let's look through a few more blogs, and it looks like everything's looking great.

[7:02] Now, let's try to create a new news area based off the blog doc types. So, for this prompt, I'm trying to get a little more specific about the doc types, and I want it to plan what it's going to do and ask me any questions before it starts. Asking it to plan adds a QA step, and it seems to work out better when it has a plan it created. This plan looks okay, and I've answered its questions. So, let's see what happens. Looks like it's done. And I'm going to check it out before I allow it to publish anything. Here's the news area. And it looks like there are content pages under it, news articles, and everything looks right. Good job again. And that's it. Now you have MCP running for Umbraco.

[7:53] Do you need help shipping MCP-powered features? We can help. Check us out at proworks.com, or just look at the link below. Thanks for watching. See you in the next video.

back to top