"author:TAM-DS" matching MCP servers:

Enables users to run and inspect local prompt-regression evaluations over six labeled fictional outreach cases, comparing a baseline and candidate prompt and reviewing per-case checks, failures, token usage, and stored run records. Exposes four local stdio tools (list_cases, run_demo, list_runs, get_run) for reviewing evidence before a human release decision, without granting paid live runs, file access, shell commands, or release approval.