This is a security layer for your MCP workflow that sits between Claude and your other tools. It scans incoming prompts for injection attacks before they hit your systems, automatically redacts PII from requests and responses, and logs all LLM SDK calls for audit trails. You'd reach for this when connecting Claude to sensitive internal tools or customer data, especially in regulated environments where you need both security scanning and compliance logging. It works as a transparent middleware, so you don't need to modify your existing MCP setup. Think of it as running your prompts through a security checkpoint before they touch anything important.
claude mcp add --transport stdio co.promptguard-security uvx security