Dev builds MCP-based gate forcing coding agents to wait for human text approval
paulofilip3 · reddit · 2026-10-04
Reddit user paulofilip3 released an open-source tool addressing the "detectably AI" tone in agent-generated text.
- Pain point: client emails, PR descriptions, and commit messages still read as AI-written no matter the prompt or humanizer skill.
- Approach: pairs an open-source CLI that installs hooks on commands like git commit and file writes with an MCP server that queues outgoing messages — agents must route text through MCP and wait for the human to edit and approve before it's sent.
- The author is seeking feedback on whether this matches others' workflows of copy-pasting agent drafts between terminal windows, or discovering agents pushed code with obviously AI-written commit messages.
More from coding & agent
- xAI Employees Run 50+ Grok Bots, Orchestrated by Manager Bots — petergyang · 2026-10-05
- Paper Finds Personal Agents Get Worse as Memory Notes Pile Up — rohanpaul_ai · 2026-10-05
- Dev Discovers You Can Mirror a Mac-Running Simulator Onto Your Phone — itsOmSarraf_ · 2026-10-05
- Agent Memory Has a Sweet Spot: 10 Lines Optimal, Code Beats Rules for Tracking — rohanpaul_ai · 2026-10-05
- Raven V2 Brings Agentic Modeling to Rhino/Grasshopper, 15k+ Seats — burhop · 2026-10-05
- Indie Dev Ships Three AI-Made Games in Half a Month, First Already on App Store — ezshine · 2026-10-05