Will Jamf Pro introduce the MCP model context protocol?

[XuHao]

We’re truly excited about MCP and can’t wait to explore its potential. The idea of using the MCP (Model Context Protocol) connected to Claude for managing enterprise devices through simple chats is incredibly appealing to us. If we were able to achieve similar functionality with JamfPro’s MCP, it would certainly be a fantastic solution.

[AJPinto]

I don’t think Jamf is going to roll out anything like this anytime soon. They’re working on AI in their own way, but what you’re describing would require a completely different level of integration and would come with a ton of risk.

Picture this — someone with very little experience tells the system, “Make Chrome downloads stop working on all devices,” and the AI is supposed to just figure it out. It would need to know how to build a config profile, pick the right domain and key pair for Chrome, apply the settings correctly, and set the right scope. All of that without testing or even confirming what the admin actually meant. That’s a recipe for disaster.

Same with something like a help desk tech asking the system to reboot a machine. If it restarts the wrong one, or hits a group of devices by accident, that’s not just inconvenient — it’s a real problem.

If this kind of functionality is something you're excited about, I’d recommend reaching out to Claude. Jamf’s API already supports most of what you’d need, so the real question is whether Claude can be trained to interact with it in a meaningful way. I’ve seen some great results using Jamf with platforms like ServiceNow, but it always takes a lot of work behind the scenes to build out that automation in a safe and controlled way.

[XuHao]

What you mentioned is indeed a very serious concern. When an inexperienced administrator sends an instruction to the system, the system may not be able to discern whether the administrator's true intention aligns with the instruction or if it was mistakenly issued. If an incorrect command is executed, the consequences could be severe and unacceptable for everyone involved.

However, let’s consider this scenario: If we were to place certain restrictions on the functions of MCP—such as limiting its capabilities to generating reports, troubleshooting issues, offering suggestions, and issuing system warnings—you could simply ask the AI, "There’s a macOS device that can’t install or publish applications. Could you please gather the relevant logs and analyze them for me?" The AI would promptly respond, informing you that the issue might stem from the device’s system version being too low to support the software.

Alternatively, you could request, "Could you help me calculate the usage duration of Adobe software across all devices and break it down by department and region? Once summarized, could you generate a Word analysis report for me?" Then, while the AI processes your request, you could take a moment to relax and enjoy a cup of coffee. By the time you return, you’ll find a comprehensive usage analysis report ready for you to share directly with your leader.

[McAwesome]

God I hope not.  Knowing that generative AI slop is being forced into every product and introducing new and unique vulnerabilities is already a nightmare.  Being expected to use something like Claude to do my job securing devices would make my industry significantly worse.

[XuHao]

If the arrival of AI is inevitable, then what we can do is to get acquainted with it in advance and learn how to make the most of its potential. At this stage, I would say that AI is still in its early development phase. While it has its limitations, such as providing inaccurate information or confusing results at times, we can work together to improve it and unlock its full potential.

[McAwesome]

It's "inevitable" just like how the Stanley Steamer was the inevitable future of the automobile, NFTs were the inevitable future of art, and how it was inevitable that we'd all have all of our banking activity on the Blockchain by 2024.  Anyone who tells you some technology is inevitable is either an idiot or thinks you're an easy mark.

[MatG]

I wrote my own Jamf MCP server to query Jamf so you can ask questions

It’s Experimental so don’t expect too much

https://github.com/mat-griffin/jamf-mcp

[XuHao]

That's absolutely fantastic! I've already had a successful experience trying it out. Thank you so much for introducing this possibility to me—it’s truly appreciated!