Building security tooling for AI agents — MCP server reputation checks, prompt-injection-sourced breach detection, and identity risk scoring for autonomous agent identities.