Quandri argues MCP is dead due to context bloat, reliability issues, and overlap with CLI/API, advocating CLI-first or Skills as the practical alternative; it backs this with measured context usage and performance data, plus a note that newer tooling (Tool Search with Deferred Loading) mitigates some issues.
Caveman is a Claude Code skill that dramatically cuts token usage (~75%) while preserving technical accuracy, with install guidance, usage triggers, and active repo metrics (MIT license, stars/forks, commits).
Mcp2cli claims a universal CLI for APIs with 96-99% token reduction versus native MCP.