Geo Guard AI — AI skill for Claude Code

geo-guard-ai

[](https://www.npmjs.com/package/geo-guard-ai)

**English** · [Русский](./README.ru.md)

**Geo-restriction for AI CLIs.** Lets Claude Code (or any other command) start only if your external IP resolves to an allowed country. If you're not where you should be, it blocks you before a single prompt goes out.

Also gates the **Cursor** chat (IDE and `cursor-agent`) the same way, via its own hook config — see [Cursor](#cursor) below.

Cross-platform: **macOS / Linux / Windows**. TypeScript, runtime — Node **18.20+ or 20.12+** (not 18.0–18.19, not 19.x, not 20.0–20.11: those are the versions where `spawn` refuses to launch a `.cmd` on Windows).

The CLI speaks **English or Russian**, picked automatically from your machine locale (`LC_ALL` / `LC_MESSAGES` / `LANG`), English by default. Force it with `GEO_GUARD_LANG=en|ru`.

---

Why

Sometimes you're only allowed to use an AI tool from a specific country — company policy, a client's terms, jurisdiction, or a personal rule of "I don't work from the wrong place." The catch is how easy it is to forget: the VPN drops, you travel, the network switches — and you just keep working as if nothing happened.

`geo-guard-ai` is **insurance against "accidentally kept working from the wrong place."** It checks your country by external IP at two points:

Checkpoint	What it does
Command start	typing claude reaches a small shim on your PATH that checks the country first, and launches the real Claude Code only if it's allowed. The same goes for cursor-agent
Every prompt	The Claude Code UserPromptSubmit hook calls geo-guard check before each prompt is sent; country not allowed → the prompt is blocked (exit 2)

The second checkpoint matters: you can start a session from an allowed country, and an hour later the VPN drops — the hook catches it on your next prompt.

Coverage

Where	Gated by
Term