Hi HN — I built Runiq: a local “sovereign runtime” that lets LLM agents (Claude Desktop, local Llama, etc.) use hardened tools to interact with your OS, instead of executing arbitrary shell/Python.

Runiq implements Anthropic’s Model Context Protocol (MCP) and exposes a small set of controlled capabilities:

Stealth Chromium browser for scraping/automation (aimed to survive common bot checks)

File operations and local workflow tools

Human-in-the-loop security: native OS popups block risky actions (write/delete/etc.) until you approve

It ships as a single static Go binary (no venv/daemon). It’s localhost-only and auditable; if you pair it with a local model, nothing leaves your machine. There’s also an optional Python adapter to use the same tool interface with OpenAI/Groq/LangChain-style agents.

Repo: https://github.com/qaysSE/runiq

I’d love feedback on the MCP implementation in Go, and on the security/UX tradeoffs for permission-gated agent actions.