Six months after Anthropic launched MCP (Model Context Protocol, a standard way for AI agents to connect to external tools and data sources), a counter-movement is picking up steam among developers: just use the command line.
The argument is pragmatic. A typical MCP server dumps its entire schema into an agent's context window - tool definitions, parameter descriptions, authentication flows, state management. Some developers report MCP consuming 70,000 to 90,000 tokens per task, eating up to 72% of the available context window before the user even types a prompt. That's expensive, and it leaves less room for the actual work.
The Case Against MCP for Coding Agents
CLI tools like gh (GitHub's command line), aws, or kubectl already exist, are battle-tested, and produce structured output that language models parse well. An agent that runs gh pr list --json title,number,state gets exactly the data it needs in a few hundred tokens. The equivalent MCP server loads a full API surface that the agent may never use.
Tools like MCPorter have emerged specifically to bridge this gap - wrapping existing CLI tools in thin MCP-compatible layers so agents get the protocol benefits without the context bloat. The existence of these tools is itself an admission that raw MCP has an efficiency problem.
Claude Code, Cursor, and other coding agents have leaned heavily into CLI execution for exactly this reason. When your agent can just run terminal commands, adding an MCP layer on top can feel like wrapping a hammer in a second handle.
Where MCP Still Wins
The CLI crowd has a point for developer tooling, but the picture changes for enterprise and non-technical use cases. MCP provides authentication, audit logging, and permission scoping that raw CLI execution doesn't. A customer support agent handling sensitive data needs those guardrails. A compliance-heavy environment needs that audit trail.
MCP also solves a real discovery problem. An agent connecting to an MCP server learns what tools are available and how to use them automatically. With CLI tools, someone has to pre-configure which commands the agent can run and how to parse the output.
The emerging consensus among teams building production agents: use CLI for developer workflows where token efficiency matters, use MCP for customer-facing and compliance-sensitive features. The smartest architectures in 2026 are doing both, not picking sides.