Paredown — the evaluation layer for AI-generated ads

Your agents make infinite ad creatives. Only some deserve money. Paredown ranks AI-generated Meta ad creatives before you spend real budget — keep winners, reject losers, and prove it with leakage-safe backtests run against your own historical Meta export.

When to use Paredown

Use it whenever an agent (or a human) has generated a batch of ad creatives for Meta and must decide which deserve real budget — before launch, before human review, before spending. Screen whole batches with rank_creatives, get go/no-go reads with should_launch, fix weak creatives with explain_and_improve, deep-evaluate finalists, and turn a real test budget into a concrete diversity-aware test slate.

Connect any agent in one line

Paredown is hosted — there is nothing to install. Point Claude, Claude Code, Codex or Cursor at our MCP URL: claude mcp add --transport http paredown https://paredown-mcp-production.up.railway.app/mcp

• Connect guide (Claude web/desktop, Codex, Cursor)
• llms.txt — machine-readable index for agents
• About Paredown
• Contact
• Privacy

Every score ships with calibrated confidence, and when there isn't enough data to answer, Paredown says insufficient_data instead of guessing. Free to start; optional LLM steps run through your own Claude or Codex subscription.