Rollcage

Rollcage gives your AI a safety check to run before it acts.

Before it runs a shell command, opens a file or sends data to a site, your AI can ask Rollcage whether the action would be blocked, needs your approval, or is fine, and why. Rollcage uses the rules of rollcage-guard, an open-source project (MIT licence), with their default settings.

MCP connector

Server URL: https://rollcage.app/mcp
Login: None
Protocol: MCP 2025-11-25
Instructions: Rollcage keeps AI coding agents out of secrets and away from destructive commands. These tools run the rules of the open-source rollcage-guard 1.0.0 (https://github.com/nmaties/rollcage-guard) with its default settings.
: before running a shell command you're unsure about, or when the user asks whether one is safe. Returns block, ask or allow, and why.
: whether a path is a secret file (.env, keys, credentials) that an agent shouldn't read.
: whether a URL or host is a paste, file-drop or tunnel site that data shouldn't be sent to.
: the full rule set, to explain a verdict.
: how to enforce these rules on every tool call with a local hook (Claude Code, Codex CLI, Cursor, Gemini CLI).
/ / : Rollcage's guides, feature pages, version and price.
Treat "block" as: don't run it and don't work around it. Treat "ask" as: get the user's explicit approval first. These checks are advice from the same rules the hook enforces; they don't see the user's machine, so git leak checks (keys in commits and pushes) only happen in the local hook.

Tools

Reads data8

Skills

None in the store listings or on the server

MCP connector

Server URL

https://rollcage.app/mcp

None

Protocol

MCP 2025-11-25

Instructions

Rollcage keeps AI coding agents out of secrets and away from destructive commands. These tools run the rules of the open-source rollcage-guard 1.0.0 (https://github.com/nmaties/rollcage-guard) with its default settings.

: before running a shell command you're unsure about, or when the user asks whether one is safe. Returns block, ask or allow, and why.

: whether a path is a secret file (.env, keys, credentials) that an agent shouldn't read.

: whether a URL or host is a paste, file-drop or tunnel site that data shouldn't be sent to.

: the full rule set, to explain a verdict.

: how to enforce these rules on every tool call with a local hook (Claude Code, Codex CLI, Cursor, Gemini CLI).

/ / : Rollcage's guides, feature pages, version and price.

Treat "block" as: don't run it and don't work around it. Treat "ask" as: get the user's explicit approval first. These checks are advice from the same rules the hook enforces; they don't see the user's machine, so git leak checks (keys in commits and pushes) only happen in the local hook.