In current AI Agent development, we are used to feeding context to Agents using Skills, Tools, MCP, or Prompts.
As agent tools continue to evolve, the initial hype around MCP has cooled down. While it standardizes how APIs are called, it also makes things more complex. It requires clients that support its protocol, and the server needs a matching protocol implementation. If it's just wrapping a bunch of API endpoints, why not just write code to call those APIs directly? Skill is a better alternative for this. More importantly, it provides instruction manuals that models can easily understand....