Claude Ads docs: how to audit your ad accounts.

No. The skill runs locally in Claude Code. If you opt into a live API connection, data flows directly from your machine to the platform API, never through us. There is no Claude Ads account, login, or server.

No. By default Claude Ads analyzes data you provide: platform exports, screenshots, or pasted metrics. It never asks for your ad-account credentials and does not connect to platforms on its own. If you configure read-only API access, the connection runs directly between your machine and the platform API, under your control.

Twelve: Google Ads, Meta, YouTube, LinkedIn, TikTok, Microsoft Advertising, Reddit, Snapchat, and X for search, video, and social, plus Apple Ads, Amazon Ads, and Pinterest for commerce and retail media. Run /ads audit to check them in parallel, or target one at a time with its shortcut such as /ads google or /ads meta.

Yes. Claude Ads is MIT-licensed and fully forkable: you can read, edit, and white-label every skill and every report. There is no per-seat fee, no credit card, and no paid tier of the plugin itself. The only requirement is a Claude Code subscription, since everything runs inside Claude Code on your machine.

Not silently, and never by default. Claude Ads is read-only out of the box. Version 2 can draft campaign launches and optimizations with /ads launch --draft and /ads optimize --draft. Applying a draft requires your explicit approval through capability-gated adapters with verification and rollback. Permanent deletion is not supported.

Every finding carries an explicit confidence level (high, medium, low, or none) tied to its evidence source, from platform API data down to flagged assumptions. Health scores are 0-100, deterministic, and weighted by spend, so a broken pixel on your biggest platform hurts more. Evidence coverage is tracked separately: 80% or more coverage gives a graded score, 60-79% is provisional, and below 60% the run is marked insufficient rather than pretending to be complete.

Three things: it runs inside your own Claude Code session with no Claude Ads server behind it; it is MIT-licensed and forkable so you can read and edit every skill; and every claim it makes is tied to dated evidence with an explicit confidence level instead of a black-box score.

Yes. The MIT license allows it, and you can white-label the reports produced by /ads report. Run it in the first 48 hours of a new client engagement to benchmark the account before you take over.

Install the Claude Ads plugin in Claude Code, point it at your account (paste an export or configure read-only API access), and run /ads google. You get an evidence-backed health score and a prioritized fix list in minutes, free and MIT-licensed. The manual equivalent costs a senior analyst most of a day.

Yes. Claude Ads flags broad-match leakage, missing negative keywords, untracked conversions, and redundant targeting, then ranks each finding by severity and evidence so you fix what costs the most first. It is free, open-source, and runs locally; the only requirement is a Claude Code subscription.

Claude Ads adds no destination for your data: the audit runs inside your own Claude Code session, under the Claude setup you already trust. There is no Claude Ads server, account, or login. And because findings are graded against a deterministic scoring contract, you get a consistent audit every run, not a different answer each time you rephrase a prompt.

Yes: Claude Ads. It is MIT-licensed and forkable, runs locally with no account access, and covers 12 platforms with confidence-rated, evidence-backed findings. There is no monthly subscription, and you can read and edit every skill yourself.