The article critiques the ongoing debate between using MCP (Managed Context Protocol) and CLI (Command Line Interface) for interacting with LLMs (Large Language Models). The author, a developer at Sentry, explains that many users misunderstand how context operates within these systems. After launching a new CLI, the common response was a desire to abandon MCP, which the author argues misses the point of MCP's capabilities. MCP's strength lies in how it steers agents and exposes tools through standardized authentication, rather than serving merely as a proxy or replacement for existing APIs or documentation. MCP is described as complex and sometimes unwieldy, but its value comes from how it helps retrieve resources efficiently. The author provides examples of the MCP's functionality, showing how it can fetch various Sentry resources like issues, traces, and profiles based on user input. This capability allows users to obtain structured data in a meaningful way, which is not easily replicated by CLI systems. The author acknowledges that CLI can perform similar tasks but emphasizes that MCP's design inherently supports ongoing context, making it more effective for certain interactions with LLMs. In discussing the CLI, the author notes that while steering context can be provided, it often requires manual intervention, which can lead to inefficiencies. The article highlights the importance of maintaining context continuously within the MCP, which contrasts with the more fragmented approach often seen in CLI implementations. The author argues that the real advantage of MCP lies in its ability to facilitate seamless interactions based on contextual information, ultimately leading to more reliable outcomes when managing tasks with LLMs.