A day on MCP: learning the concept, then turning it into a blog post
Studied what MCP is and where it fits versus a plain CLI, then drafted a bilingual post explaining the MCP client-to-server flow with diagrams. Plus interview prep.
A day that started as study and ended as published thinking. The thread tying it together was the Model Context Protocol.
tgmarinho-ai-website
I spent the morning getting MCP straight in my own head: what it is, when it actually helps, and when a plain CLI is the better tool. The interesting part is the trade-off. A CLI gives an agent a fixed set of commands; an MCP server exposes typed tools an agent can discover and call at runtime, and it is not limited to AI clients. That distinction is what most explanations skip.
Once it clicked, I turned it into a blog post: "MCP vs CLI". It covers what MCP is, real use cases, when not to reach for it, and the difference between calling tools through MCP versus shelling out to a CLI. I drew flow diagrams of the full path, MCP client to MCP server to the tool calls, using one of the existing interactive RAG posts as a structural reference.
Learning notes
Before writing anything, I worked through the basics out loud: RBAC versus ABAC for access control, and then MCP from the ground up, including a concrete hypothetical of who the client and the server would be in a real product and how something like an event organizer's tool would connect.
career
Polished interview prep and tightened a resume layout for printing, keeping the wording honest about past impact.
Teaching a concept is the fastest way to find the holes in your own understanding, and the post exists because the study forced them into the open.