MCP server: security scanner for AI-generated code — the agent scans its own code/diff before committing. Detects injection, SSRF, hardcoded secrets, weak crypto, unsafe deserialization, TLS-off, XSS. Deterministic, free, no LLM.