Agent-first

Connect paredown to your agent

One hosted URL. No account, no install, no card. Your agent gets rank / explain / deep-evaluate / test-slate tools for AI-generated Meta ads.

The URL

https://paredown-mcp-production.up.railway.app/mcp

Everything below is just different ways of handing your agent that URL.

Door 1 · Let the agent do it

Paste this into any coding agent (Claude Code, Codex CLI, Cursor):

Add the paredown MCP server at https://paredown-mcp-production.up.railway.app/mcp
(streamable HTTP). Then confirm it works by calling evaluation_info.

Or run it yourself:

claude mcp add --transport http paredown https://paredown-mcp-production.up.railway.app/mcp

Door 2 · claude.ai web & desktop

Settings → Connectors → Add custom connector
Name: paredown
URL: https://paredown-mcp-production.up.railway.app/mcp
Ask Claude: “Rank these 20 ad creatives before I spend budget.”

Door 3 · Manual config file

{
  "mcpServers": {
    "paredown": {
      "url": "https://paredown-mcp-production.up.railway.app/mcp",
      "transport": "http"
    }
  }
}

What your agent can do afterwards

rank_creatives — screen a whole batch of generated creatives before spend
should_launch — quick go/no-go on one creative
explain_and_improve — concrete revisions for weak creatives
deep_evaluate + recommend_test_slate — finalist ensemble → real test budget split
backtest_account — prove it on your own Meta history first

Every answer carries confidence and an honest insufficient_data when there isn't enough to go on — paredown never fakes certainty.