MCP. Directory has published a detailed comparison of two of the most popular agent-based CLIs for programming — Claude Code by Anthropic and Codex by OpenAI.
The authors conclude that there is no overall winner between them: both tools support MCP, can independently edit code, work with Git, run commands, and use subagents.
The key factors of choice are the model on which the agent works and the subscription already paid by the user. At the same time, the scenario of using both CLI simultaneously is becoming more and more popular among experienced developers: one agent writes code, and the other independently checks changes before committing.