Cordon

Security proxy that enforces granular policies and audit logging for AI agent tool access.

Acts as a governance layer between AI agents and MCP servers, intercepting tool calls to enforce allow/block/approve policies per tool or server. Maintains immutable audit logs for compliance, supports human-in-the-loop approval via terminal or Slack, and prevents unrestricted agent access to databases and APIs. Written as a TypeScript monorepo with CLI, SDK, and proxy engine packages.