"Audit your current MCP servers & ask, for each tool they expose, whether that tool solves a knowledge problem or an execution problem. If a tool exists primarily to teach the agent how to use an API rather than to call that API, it is a candidate for extraction into a skill file." https://thenewstack.io/skills-vs-mcp-agent-architecture/
The case for running AI agents on Markdown files instead of MCP servers

Developers are replacing bloated MCP servers with Markdown skill files — cutting token costs by 100x. Here's the two-layer architecture emerging in production AI systems.

The New Stack