HOL Guard

HOL Guard is a local-first runtime firewall for AI coding agents. It intercepts tool calls — shell commands, secret access, MCP tools, and package installs — before they execute on your machine, so you can approve, deny, or audit risky actions in real time. Unlike cloud scanners or static policy packs, Guard runs locally next to the agent runtime and enforces controls at the moment of tool use. Install with: pipx install hol-guard && hol-guard init. HOL Guard is Apache-2.0 open source from maker HOL. It is free to use. Scan is not a safety guarantee — Guard is a runtime control and review layer, not an antivirus and not a promise that every agent action is safe. Primary audience: developers and security teams who ship AI coding agents (Cursor, Claude Code, custom MCP setups) and need local visibility and policy before agents touch the shell, secrets, or the package manager. Core capabilities include tool-call interception for shell, secrets, MCP, and installs; local-first operation with no requirement to upload your workspace to a SaaS scanner; honest defaults that fail closed on dangerous actions when configured; and a simple init flow. Website: https://hol.org/guard. Windows desktop MSI is also available for teams that prefer a packaged client. Use HOL Guard when you want runtime protection for agent tool use without giving up local control. Do not treat a clean scan or an allow decision as a warranty of safety; always review high-impact actions. Maker: HOL. Free. Apache-2.0. HOL Guard focuses on practical local runtime controls for everyday agent workflows across IDEs and CLIs.