Running Claude Code means giving an agent access to your repo, tools, and runtime.
This guide shows how to keep control: local models, MCP tool access, and sandboxed execution working together.
The combo that makes agents usable in real workflows, not just demos:
顯示更多