Marketing & Business · Blog Post
Claude Ads: a 12-platform paid-media audit & ops skill for Claude Code
Twelve ad platforms, one shared rule: Claude Ads by AgriciDaniel doesn't get write access to any of them until you explicitly grant it. Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat, and X all get audited the same way — read-only, source-grounded, scored against 161 to 250+ checks per audit with a four-state pass/fail/unknown/not_applicable verdict, versioned into a JSON report you can compare run over run. The nine-command pipeline (setup, audit, plan, create, launch, monitor, optimize, experiment, report) does eventually reach into a live account, but only after you flip on the specific capability that allows it and sign off on the change. Same author as this site's claude-seo review, 7,744 GitHub stars, MIT license.
By Mehadi Hasan 7 min read
What it is
claude-ads is an open-source Claude Code skill built for the messier side of digital marketing: managing paid ad accounts across a dozen networks without babysitting every platform's dashboard by hand. It's built by AgriciDaniel — the same author behind claude-seo, reviewed here previously — and describes itself as "Claude-first, portable paid-media operations for agencies, consultants, and in-house performance teams." The repo has crossed 7,744 GitHub stars, ships under the MIT license, and had its most recent push on July 13, 2026, a few months after its February 2026 creation.
What separates it from a general "write me some ad copy" prompt is the audit engine underneath. Every check resolves to one of four states — pass, fail, unknown, or not_applicable — and platform coverage gets graded on what evidence is actually available: 80%+ coverage earns a real grade, 60–79% comes back provisional, and anything under 60% is flagged insufficient rather than quietly skipped. That's a meaningfully different posture than an LLM guessing at what "good" looks like for a Meta account it's never seen data from.
Twelve platforms, nine commands
The 12 platforms split roughly into two groups: the big paid-social-and-search names — Google Ads, Meta Ads, YouTube Ads, LinkedIn Ads, TikTok Ads, Microsoft Advertising, X Ads, and Snapchat Ads — plus a commerce-and-retail set that gets less tooling attention elsewhere: Apple Ads, Amazon Ads, Pinterest Ads, and Reddit Ads. Shortcuts exist for the busiest ones (/ads google, /ads meta, /ads amazon, /ads reddit), so a single-platform check doesn't require invoking the full 12-network sweep.
The command set runs as one pipeline rather than nine disconnected utilities: /ads setup configures the client, account, KPI, and privacy profile once, then /ads audit → /ads plan → /ads create → /ads launch --draft → /ads monitor → /ads optimize --draft → /ads experiment → /ads report carries a campaign from evidence-gathering through to a validated report bundle. Two more commands sit alongside that core flow — /ads research refresh to pull updated platform evidence, and /ads validate to check contracts and readiness.
Installation
Claude Code plugin marketplace
/plugin marketplace add AgriciDaniel/claude-ads /plugin install claude-ads@ai-marketing-hub-claude-ads
That's the canonical path inside Claude Code; run /ads setup right after to configure your first client/account profile. If you'd rather skip the plugin marketplace, cloning the repo and running bash install.sh --source=local works identically, and the same install script also targets Codex, Gemini, Cursor, Windsurf, and Goose via --target=<host> flags. It needs Python 3.11 or 3.12; Playwright is optional for browser-based capture, and WeasyPrint plus Pango are optional if you want PDF report renders on top of the default JSON, Markdown, and HTML output.
Using it day to day
Day to day, the loop is audit first, act second. Say you're running paid media for a mid-size DTC brand spending around $18k/month split across Google Search and Meta — you'd run /ads audit google and /ads audit meta ahead of a monthly review, and get back a graded report where every line is a specific check (naming conventions, conversion tracking health, audience overlap, budget pacing against KPI) marked pass, fail, unknown, or not_applicable, not a vague "your account could be better optimized" summary. The versioned JSON output means you can diff this month's audit against last month's and see exactly what regressed.
The write side stays deliberately slower. /ads optimize --draft on an underperforming ad set doesn't touch anything live — it produces a human-readable diff of the proposed bid or budget change, which needs the capability explicitly enabled, explicit IDs, and your sign-off before /ads launch actually applies it, complete with an idempotency key and a rollback window if the change needs to be undone.
My take
The honest read: this is a serious tool for someone who already knows what a good paid-media audit looks like and wants the grunt work of running one, consistently, across 12 platforms, automated — not a tool that replaces the judgment call of whether a 40% ROAS drop means kill the campaign or wait out a seasonality dip. The four-state scoring and coverage grading are a genuinely good design choice; an "unknown" result because a platform didn't expose the data is a more honest answer than an LLM confidently inventing a pass or fail.
This is AgriciDaniel's second tool covered here — after claude-seo's falsifiable-audit approach to organic search, the same evidence-first, capability-gated design shows up again here, just pointed at paid media instead. If you're running ads as one channel among many and don't need per-platform depth, aaron-marketing-skills has a paid-ads discipline with its own ad-account-auditor that covers the basics; reach for Claude Ads specifically when you're managing spend across several of these 12 networks and need audit depth per platform, not just a general pass.
Frequently asked questions
Does Claude Ads need write access to my ad accounts to run an audit?
No. It's read-only by default — audits, plans, and reports run without any live capability enabled, and the system's own design principle is blunt about it: "missing ceilings mean no write." You only get write access to a platform once you explicitly enable that capability.
Which 12 ad platforms does it actually cover?
Google Ads, Meta Ads, YouTube Ads, LinkedIn Ads, TikTok Ads, Microsoft Advertising, X Ads, Snapchat Ads, Apple Ads, Amazon Ads, Pinterest Ads, and Reddit Ads. Platform shortcuts exist for Google, Meta, Amazon, and Reddit for faster single-network commands.
Can it make changes to a live campaign on its own?
Only after you explicitly turn on write capability for that action. Even then, changes go through a human-readable diff, explicit resource IDs, owner approval, an idempotency key, a check that remote state still matches what the diff assumed, and a rollback window.
What does it cost, and what do I need to run it?
The code and docs are MIT-licensed and free to self-host. You need Python 3.11 or 3.12 on Linux, macOS, or Windows; Playwright is an optional add-on for browser-based data capture, and WeasyPrint plus Pango are optional if you want PDF exports alongside the default JSON, Markdown, and HTML reports.