The piece breaks down the concepts of MCP (Multi-Context Protocol), Skills, Commands, and Agents in coding environments. It challenges the recent hype around Skills, which are presented as reusable prompts that enhance an agent's capabilities. Skills can bundle scripts and other artifacts, allowing agents to perform specific tasks like creating pull requests or simplifying code. The author argues that while Skills are useful for encoding routine tasks, MCP is often misunderstood and unfairly criticized. MCP connects tools and agents, allowing for the exposure of remote procedure calls (RPCs) as tools. The article emphasizes that while Skills are about enhancing an agent's abilities, MCP serves as a framework to expose and manage those tools effectively. Poor implementations have tarnished MCP’s reputation, leading to an overflow of unnecessary tools and inefficient use of context tokens. Despite these issues, MCP's strengths lie in its OAuth-native authentication and the flexibility it offers organizations to control which tools are available to users. The author highlights that MCP can improve context steering and response crafting, offering more tailored interactions. For instance, Sentry’s MCP service provides specific guidance alongside issue details, optimizing user experience. While some MCP servers may be redundant or poorly designed, the protocol itself holds significant potential for structured and efficient tool management in software development.