MCP's total victory: why MCP beats CLIs for most agent integrations
teropa · x · 2026-09-16
Rhys Sullivan declares "total MCP victory," arguing MCP beats CLIs for most integrations: indexable tool catalogs that scale to unlimited tools, no full-sandbox requirement, consistent auth across all MCPs, multi-account support, and code-mode implementations that let the model know what will be returned for efficient token usage.
He attributes MCP's slow adoption to bad client implementations — no hot reloading (you had to restart the whole client) and agents being less familiar with debugging MCPs than CLIs. Over the past year, codex and Claude plugins have all used MCP under the hood, and computer use is essentially an MCP.
The quoted thread adds that better tool calling, tool deferral, and MCP going stateless enabled the shift; tip: add params like query to MCP tools for composing/filtering data.
Related event: Developers Flip Toward MCP Over CLI for Agent Tool Integration(5 posts)→
More from coding & agent
- Dev builds Sturzi, a Grok bot that finds and files simpler versions of Argentine regulations — StewartalsopIII · 2026-09-16
- Compaction vs prompt caching: subagent context isolation may beat summarizing in production agents — Marcus_MSC · 2026-09-16
- Sakana Marlin adds Interactive Reading and PowerPoint export for deep research reports — SakanaAILabs · 2026-09-16
- Andy Matuschak demos programmable highlighters that trigger AI actions on annotations — andy_matuschak · 2026-09-16
- Dev shares trick: dedicated subagent filters SSH output to protect agent context — SaleemUsama · 2026-09-16
- Raw Markdown Folders Kill Your Agent's Context Window: Lessons From 100k Installs — Alone-Biscotti6145 · 2026-09-16