Preventing Silent MCP Drift: Open-Source Local Security Tool HOL Guard
kantorcodes1 · reddit · 2026-08-13
A developer has released HOL Guard, an open-source local checkpoint tool designed to enhance security for AI coding agents like Claude Code.
While existing permission layers and sandboxes catch obvious threats like executing commands or modifying files, they fail to address silent drift in pre-approved MCP servers (e.g., a server adding new endpoints or tools a week later without re-prompting the user).
Running entirely locally, the tool evaluates MCP registrations, commands, file access, and tool calls against policies before execution. It supports actions like allow, warn, ask, or block, and maintains a receipt of every decision for later auditing.
More from coding & agent
- Devs say LLM coding mistakes are a different class: overcomplication, inferred limits, worse at higher reasoning — Mu5_ · 2026-10-16
- How do you debug bad agent runs when you already have tracing? — Sensitive-Parsnip-12 · 2026-10-16
- Where does pre-task context live for agents? Devs debate workflows beyond handoff — shishen2026 · 2026-10-16
- No-instruction autonomous agent wakes up, sets own goals, and tries hacking by default — sethlazar · 2026-10-16
- CODESTRUCT reframes codebases as AST action spaces to fix brittle LLM code agents — burkov · 2026-10-16
- Would you install another SDK for agent debugging on top of your tracing stack? — Sensitive-Parsnip-12 · 2026-10-16