When does MCP make sense vs CLI? (ejholmes.github.io)

The Model Context Protocol (MCP), once hailed as a revolutionary way for large language models (LLMs) to interact with services, is facing critiques regarding its practicality and necessity. The article argues that MCP’s attempts to create a standardized and cleaner interface for LLMs are misguided, as these models can already adeptly utilize command-line interfaces (CLIs) to execute commands and access documentation. Instead of promoting efficiency, MCP introduces unnecessary complexity—such as authentication issues and operational friction—compared to the straightforwardness of traditional CLI tools. For the AI/ML community, the significance of this perspective lies in the reminder that existing technologies may already fulfill user needs efficiently. The article emphasizes that CLIs support a composable, debuggable framework for both humans and machines, benefiting from years of real-world refinement. By advocating for prioritizing solid APIs followed by robust CLIs, the piece calls on AI developers to reevaluate their focus on new protocols like MCP, suggesting they return to leveraging proven tools that streamline interactions between LLMs and services without additional burdens.