Meta Ads Auditor

A Claude Code skill that runs a weekly health audit on a Meta ad account — straight from Claude, on Meta’s own native ads MCP server. No API keys, no CSV exports, no clicking around Ads Manager.

Point it at an account and it:

1. Pulls the last 7 days + the prior week (so every metric has a trend).
2. Auto-detects the objective — grades on ROAS/CPA for sales accounts, CPL/CPC/CTR for lead-gen accounts.
3. Runs 18 health checks across efficiency trend, creative fatigue, wasted spend, winners to scale, delivery hygiene, and structure.
4. Scores the account 0–100 with a week-over-week delta.
5. Returns the 3 things to fix this week — ranked by severity, each with the number behind it and the exact move — plus the one winner you should scale.

It’s read-only: it tells you what to do, it doesn’t touch your budgets or ads.

  META ADS AUDIT · Northwind Supply Co. · last 7 days   [objective: SALES]
  ────────────────────────────────────────────────────────────────────────
  HEALTH SCORE   68 / 100   ▼ from 79 last week

  🔴 FIX THIS WEEK
  1. Spend up 22%, returns down 16% — you're scaling into weakness.
  2. Creative fatigue: "UGC – Sarah Testimonial" CPA +38% ($22→$30) at freq 5.1.
  3. $3,120 on 4 zero-purchase ads → kill list.

  ✅ SCALE THIS: "Problem-Solution v3" holding 4.6 ROAS, capped at $80/day.

  ✓ 14 passed   ⚠ 3 flags   ✕ 1 critical

Setup (one time)

Connect Meta’s native ads MCP server to Claude:

• Server URL: https://mcp.facebook.com/ads
• In Claude / Claude Code, add it as a remote MCP (custom connector).
• Manage what the agent may do per account in Business Suite → Settings → Integrations → Ads MCP server. All write actions are paused-by-default on Meta’s side; this skill only reads.

Then drop this folder into ~/.claude/skills/meta-ads-audit/ and say “run my weekly meta ads audit.”

Files

• SKILL.md — the skill (methodology + report spec).
• references/audit-checks.md — the 18 checks and the scoring rubric.

Notes

• Built 100% in Claude Code.
• Read-only by design. Acting on a fix is a separate, explicitly-authorized step.
• Numbers-first: every finding carries the metric behind it, or it doesn’t ship.

Built by

Built by Mike Futia. For more Claude Code skills that help DTC brands and creative agencies run on AI, join SCALE AI → https://www.skool.com/scale-ai/about