Hey everyone, I just finished a project that might interest you.
An MCP server which consumes the cache dir of a checkmk instance. It gives you the possibility to answer questions like:
which additional processes should I monitor?
Since I always wanted to try out video producing, I made a YouTube video about it yesterday.
Video creation is pretty time consuming, but I learned a lot! It’s definitely not a perfect video, but I made it for you guys.
Great project! I’m genuinely interested to see how your journey unfolds.
Checkmk already collects a wealth of valuable data that often remains underused. It’s almost ironic that we need an MCP to uncover some straightforward correlations.
The real added value will emerge once these assistants address more complex questions like “What caused the outage in Service A?”, especially when the underlying issue originates from a completely different service monitored at another site or tool.
Cross-system relationships are often difficult to identify without proper autodiscovery. Tools like Dynatrace clearly demonstrate their strengths in this area, as they collect data across multiple layers and systems, then connect it in a coherent and practical manner. This makes it possible to clearly understand which service depends on which resource, and that is precisely where I see great potential for further AI development, especially since Checkmk is still blind on this eye.
Thanks, and yeah, you’re right — that’s the hard part. Right now the MCP basically only indexes by hostname, so it’s just giving fast ad-hoc access to what Checkmk already knows, not inferring cross-system relationships.
To get anywhere near Dynatrace-style dependency mapping, you’d need to process the raw Checkmk data into a proper graph/relational store — topology, dependencies, correlations over time. That’s a real ETL pipeline, not a thin API wrapper, and it’s a lot of ongoing compute for what’s still a “maybe” use case right now. Definitely interested in the idea, just being upfront that it’s not a small lift.