Last updated: October 8, 2026. Facts checked against the project's GitHub repository and documentation.
TL;DR
Claude Ads is a free, open-source Claude Code plugin that audits paid advertising accounts across 12 platforms, including Meta, Google Ads, TikTok and LinkedIn. It is an independent project by developer Agrici Daniel, published under the MIT license, and it is not an Anthropic product. It is read-only by default, and while v2 can apply some changes after explicit approval, it does not run or optimize your campaigns continuously.
Definition: Claude Ads
Claude Ads is an independent, open-source Claude Code plugin that audits paid-ad accounts and drafts recommended fixes. It is not made or endorsed by Anthropic, and it diagnoses your ads rather than operating them day to day.
Claude Ads at a glance (October 2026)
| Question | Answer |
|---|---|
| What it is | A Claude Code plugin containing a suite of paid-media Agent Skills, invoked with /ads |
| Who makes it | Agrici Daniel, an independent developer (AgriciDaniel/claude-ads on GitHub) |
| Official Anthropic product? | No. Its docs say it is "not affiliated with, endorsed by, or sponsored by Anthropic" |
| Cost | The plugin is free (MIT). Claude Code needs a paid Claude plan (Pro is $20/month or $200/year) |
| Platforms | 12, including Meta, Google, YouTube, LinkedIn, TikTok, Microsoft, Amazon and Reddit |
| Data it uses | Exports, screenshots or pasted metrics by default; live reads depend on the installed adapter and your permissions |
| Can it change campaigns? | Read-only by default. Per the README, live writes stay disabled until a specific operation passes approval, verification and rollback gates |
| Latest release | v2.0.2, September 10, 2026 |
What is Claude Ads?
Claude Ads is a paid-media audit and planning plugin that runs inside Claude Code. You type /ads audit all or a shortcut like /ads meta, and Claude works through structured, platform-specific controls instead of improvising.
It uses Anthropic's Agent Skills format: a skill is a folder with a SKILL.md file that gives Claude instructions for one job. The Claude Ads repository packages many of them: a main /ads skill, a dedicated skill and audit worker for each platform, shared references and a scoring engine. It also drafts plans, creative briefs and reports, saved as JSON and rendered to Markdown, HTML or PDF.
Dated facts matter here. According to the release notes, v2.0.0 (July 13, 2026) expanded coverage to 12 platforms, read-only by default; v2.0.2 (September 10, 2026) is the latest release. The public repository is the project's open-source mirror; its changelog says canonical development moved to a private AI Marketing Hub repository. Audit-check counts have changed across versions, and the current repository gives no fixed total.
Drowning in Meta Ads?
Put your campaign on autopilot with Nova.
Read moreIs Claude Ads an official Anthropic product?
No. Claude Ads is an independent, third-party project. The project's documentation states it is "not affiliated with, endorsed by, or sponsored by Anthropic."
Three things people mean by "Claude ads"
- The Claude Ads plugin covered here.
- Ads inside Claude. On February 4, 2026, Anthropic said Claude will stay ad-free, with no sponsored links in conversations and no advertiser influence on answers.
- Forks and look-alikes. Older copies named
claude-adsmake different claims, so confirm you are readingAgriciDaniel/claude-ads.
How much does Claude Ads cost?
The plugin is free. It is MIT-licensed, and its docs say there is "no paid tier of the plugin itself," no Claude Ads account and no Claude Ads server.
You pay for Claude Code. As of October 2026, Claude's pricing page lists Pro at $20 month to month, or $200 per year, and Max from $100/month. Claude Code is included in Pro, Max, Team and Enterprise, not in the free plan, and it shares your plan's usage limits. For how Claude's plans compare for ad work, see our Meta AI vs Claude for Meta Ads comparison.
What can Claude Ads do, and what stays human-controlled?
Claude Ads diagnoses and drafts. It does not run your account continuously. In v2, some supported changes can be applied, but only after you approve them.
What Claude Ads does and doesn't do (v2, October 2026)
| Job | Claude Ads | What that means in practice |
|---|---|---|
| Audit accounts across 12 platforms | Yes | Controls marked pass, fail, unknown or not applicable |
| Score account health | Yes | Health and evidence coverage are scored separately |
| Draft plans, briefs and experiments | Yes | Output is a draft for a human to review |
| Read live account data | Partly | Depends on the installed adapter, credentials and platform permissions |
| Apply changes | Gated | Needs a tested capability, explicit IDs, a before/after diff, owner approval and a rollback plan (per the repository) |
| Monitor and optimize 24/7 | No | It is a skill you invoke in Claude Code, not a managed always-on service |
| Optimize to margins or break-even ROAS | Not built in | Its benchmark references cover CPC, CTR, CVR, ROAS and ACOS; profit context only exists if you supply it |
Expert note: why "unknown" is the most useful word in the report
In Claude Ads, "unknown" never counts as "pass." Missing evidence lowers coverage instead of inflating the health score. Per the repository, a report with 80% or more coverage is graded, 60 to 79% is provisional, and under 60% is marked as insufficient evidence.
So interpret a health score alongside its coverage. In Meta accounts, what an export can't show tends to sit in measurement (pixel, Conversions API, attribution), where gaps can conceal performance problems. A high score with low coverage likely means the audit didn't see enough, not that the account is fine.
The Audit-to-Action Ladder
We call it the Audit-to-Action Ladder: four capability layers, each adding something the layer below lacks.
The Audit-to-Action Ladder
| Layer | Core capability | Human involvement | Example |
|---|---|---|---|
| 1. Diagnose | Find and score problems | Reviews findings | Claude Ads /ads audit |
| 2. Draft | Propose plans and changes | Approves the plan | Claude Ads /ads plan, --draft modes |
| 3. Execute | Apply specific approved changes | Approves each action or batch | Claude + a Meta Ads MCP with write permissions; Claude Ads gated changes |
| 4. Operate | Monitor and optimize continuously | Sets policies and exceptions | An AI media buyer such as Nova |
Claude Ads is strongest at Diagnose and Draft. In v2, some supported operations can reach Execute after explicit approval, but it is not a continuous operator.
For Execute, you can connect Claude to Meta through the official Meta Ads MCP or a managed connector such as the AdAdvisor MCP server, which carries business context like AOV and break-even ROAS into Claude's tools. Google's official Google Ads MCP server is read-only in its current release, so it supports Diagnose rather than Execute. See our guides to using Claude for Meta Ads with MCP and how to use Claude for Facebook Ads.
Our view: for most advertisers, the audit is not the bottleneck. Acting on it, consistently, is. A free audit tool lowers the software cost of finding problems, but not the labor of validating and fixing them.
When is Claude Ads enough, and when do you need an execution layer?
Claude Ads is likely enough if you have time to review findings and apply fixes yourself. Pair it with our AI audit process or the 40-check Meta Ads audit checklist, then act through Claude and an MCP connection.
You likely need an operating layer when the account changes faster than you can re-audit it, since delivery can shift as auction conditions, budgets and creative response change. Nova, AdAdvisor's own product, is a profit-first, approval-first AI media buyer that runs your Meta ads 24/7 inside the guardrails you set, with Iris, its creative AI manager. Per AdAdvisor's Nova page, it starts in Suggest mode, where "nothing executes until you tap approve," and uses your break-even ROAS as a guardrail in scaling decisions. It covers Meta only, so it complements a multi-platform audit. As of October 2026, AdAdvisor pricing runs from a free tier and MCP-only plans at $19.99/month to Nova at $199/month ($75 for the invite-only Founding 100).
Frequently asked questions
Summary
Claude Ads is an independent, MIT-licensed Claude Code plugin for auditing and planning paid media across 12 platforms, not an Anthropic product. It is read-only by default, with supported changes gated behind explicit approval and rollback. Its strength is structured diagnosis; continuous optimization still needs a human workflow, an MCP execution layer or an operating agent. AdAdvisor builds that last layer with more than 8 years in media buying, over $60M in managed ad spend, and an ex-Meta engineer on the team.
Sources
- Claude Ads repository, AgriciDaniel/claude-ads (GitHub)
- Claude Ads releases (GitHub)
- Claude Ads changelog (GitHub)
- Claude Ads documentation
- Agent Skills overview (Anthropic)
- Claude plans and pricing (Anthropic)
- Claude is a space to think (Anthropic, February 2026)
- Ads MCP Server overview (Meta for Developers)
- Google Ads MCP server (Google for Developers)
- AdAdvisor pricing
