But after stepping back, asking the right questions, and mapping everything to real systems I understand (Linux, APIs, Kubernetes), something clicked. This post is the breakdown I wish I had from the beginning. So the question is what MCP actually is?
What MCP Actually Is (No Hype)
After stripping away all the buzzwords, MCP comes down to one simple loop:- Give AI a goal
- Give AI a set of tools
- Let AI decide which tool to call
- Feed results back
- Repeat until done
It's not magic — it's just telling AI which function to call next. Before MCP, automation looked like this: "IF error → restart service", but with MCP it's like this:
AI sees error → decides: - check logs - analyze root cause - restart service - or escalate
The difference is static rules vs adaptive decision-making.
Honestly, it was a lot of presentations non stop which I partially recorded and took notes. I might write further blog posts in relation to the conference in another day. But one of the interesting realizations came from a talk by Hubspot about RPC. Any RPC with a schema is already a tool — which changed everything. Existing APIs in an organization is equal to MCP tools and these functions is already usable by AI. You don't need to rebuild your system.
There were other explanations such as MCP gateways and MCP registries — how it is providing enterprise IT with a point of control. Security aspect of MCP and how to make sure we are designing robust MCP app. Without doubt Docker was present and shared the MCP horror story.
I also seized the opportunity to visit some companies at the stand such as AWS, Datadog, Arcade, and Sonar. I am also in the process of building an MCP server that can take action on my own blog whenever it detects the server is going down which I will share in a different blog post.