Solving Agent Write-Access Risks: Open-Source Security Gateway 'mcpip'

Ok_Anxiety410888 · reddit · 2026-08-05

The author argues that read-only agents are easy, but granting write access (sending emails, updating DBs) exposes flaws in simple approval prompts, primarily 'approval fatigue.' To fix this, they built mcpip, a security gateway sitting in front of tool calls.

Core Design Principles:

The gateway runs deterministically (no LLM inference), speaks MCP, and is self-hostable.

Related event: Open-Source Gateway mcip Secures AI Agent Tool Calls(2 posts)→

Original post →

More from coding & agent

coding & agent channel →