"author:omnirogue" matching MCP servers:

Exposes read-only repository scanning across architecture, security, performance and product lenses, citation verification of findings against a working tree or git revision, and HTML/SARIF/Markdown dossier rendering as MCP tools for assistants such as Claude and Cursor.

Exposes offline LLM-as-judge evaluation tooling to MCP clients so they can benchmark recorded judges against golden labels, diagnose position and verbosity bias, and fit probability calibrators with agreement, cost-frontier, and rubric-template utilities. It deliberately omits command judges, keeping all analysis standard-library and network-free.