Ghostgate: a local MCP security gateway that blocks risky AI tool calls
Ghostgate sits between an AI agent and its tools, enforcing YAML‑defined policies to allow, deny or require approval for each tool call. It scores risk, redacts secrets, logs events to SQLite or JSONL, and offers a terminal approval prompt and dashboard scaffold. Designed for developers and teams building autonomous agents, it protects against accidental or malicious system access better than generic proxy or sandbox solutions.
Visit author’s GitHub →deadrootsec/ghostgate