Quick pick: for a full-site audit, install AgriciDaniel’s free claude-seo plugin (/plugin marketplace add AgriciDaniel/claude-seo, then /plugin install claude-seo@agricidaniel-claude-seo, then /seo setup) and run /seo audit <url>. For a lighter single-page check, use the seo-audit skill in Corey Haines’ marketingskills. On sites with thousands of URLs, export a crawl from Screaming Frog and hand Claude the CSV, because no skill is a crawler.
Most “AI SEO audit” posts list ten tools and stop. That’s not much use when you have a client site, a free afternoon, and a terminal open. For the wider picture beyond audits (keywords, content, schema, GEO), see Claude Code skills for SEO.
This guide is for the person actually running the audit. It covers which Claude Code skills are worth installing (free and paid), how they fetch pages, which signals they can’t see, and how to get from a 60-item findings dump to a fix list a developer will work through. Repo star counts and last-commit dates were checked against the GitHub API on 1 November 2026. Prices come from vendor pricing pages where those exist, and are marked “pricing not public” where they don’t.
What a Claude Code SEO audit skill is (and isn’t)
Short answer: a Claude Code skill for SEO audits is a folder with a SKILL.md file and, often, helper scripts. Together they tell Claude how to fetch a page or site, what to check, and how to report it. It isn’t a crawler with its own index. It runs on your machine, uses your Claude plan, and only sees what it can fetch or what you connect to it.
Skills live in ~/.claude/skills/ (available in every project) or .claude/skills/ (current project only). Some arrive inside plugins. If skills are new to you, start with what Claude Code skills are and then how to install Claude Code skills.
You’ll see three delivery methods:
| Method | Command shape | When you’ll see it |
|---|---|---|
skills CLI | npx skills add owner/repo (add --skill name to pick one) | Single skills and multi-skill repos |
| Plugin marketplace | /plugin marketplace add owner/repo then /plugin install name@marketplace | Bundles with subagents, hooks, scripts |
| Manual | git clone the repo, copy the folder into ~/.claude/skills/ | Workspaces such as seomachine |
How much an audit skill can do depends on its access. A skill that’s only instructions relies on Claude’s built-in web fetch and search. A skill that ships Python scripts and a headless browser can render JavaScript and parse the DOM. A skill wired to an MCP server (Ahrefs, SE Ranking, DataForSEO, Google Search Console) can pull ranking, backlink and crawl data that no amount of page fetching would turn up.
Takeaway: before you trust a score, find out what the skill was able to fetch.
The best Claude Code skills for SEO audits in 2026
These are the audit-relevant skills and data sources that held up under verification. Stars and dates come from the GitHub API on 1 November 2026. A star count measures popularity, not audit accuracy.
Free and open-source skills
1. claude-seo (AgriciDaniel)
- Source: github.com/AgriciDaniel/claude-seo · MIT · about 25,400 stars · last push 1 Nov 2026
- Free. Optional paid extensions (DataForSEO, SE Ranking, Profound) need those vendors’ accounts.
- What it does: the most complete free audit suite. Commands include
/seo audit <url>,/seo page,/seo technical,/seo schema,/seo geo,/seo sitemap,/seo hreflang,/seo images,/seo localand/seo backlinks. Its README says a full audit can run up to 17 agents in parallel, and that/seo setupinstalls a headless Chromium so JavaScript-heavy pages get rendered. Core Web Vitals are lab estimates unless you connect free Google APIs with/seo google setup. - Install (inside Claude Code 1.0.33+, Python 3.11 to 3.14):
/plugin marketplace add AgriciDaniel/claude-seo /plugin install claude-seo@agricidaniel-claude-seo /seo setup
2. marketingskills (Corey Haines): seo-audit, schema, ai-seo
- Source: github.com/coreyhaines31/marketingskills · MIT · about 57,800 stars · last push 31 Oct 2026
- Free.
- What it does: three skills that reference each other.
seo-auditdiagnoses SEO issues,schemaadds or fixes structured data, andai-seocovers getting cited in AI answers. They’re lighter than claude-seo because they’re mostly instructions, which also makes them easy to read and edit. - Install:
npx skills add coreyhaines31/marketingskills --skill seo-audit ai-seo schema(inside a Claude Code session, add-a claude-code).
3. Anthropic knowledge-work-plugins: Marketing seo-audit
- Source: github.com/anthropics/knowledge-work-plugins · Apache-2.0 · about 34,300 stars · last push 1 Nov 2026
- Free (official Anthropic).
- What it does: lives under
marketing/skills/seo-audit. It covers keywords, on-page elements, content gaps, technical issues and competitor comparison, and outputs a plan split into quick wins and strategic work. With an SEO tool connected over MCP it uses real ranking and volume data. Without one, it falls back to web search and estimates, which you should treat as directional. - Install:
npx -y skills add anthropics/knowledge-work-plugins --skill seo-audit --agent claude-code
4. claude-skills (Alireza Rezvani): marketing bundle
- Source: github.com/alirezarezvani/claude-skills · MIT · about 29,100 stars · last push 31 Oct 2026
- Free.
- What it does: a large multi-domain library. Its marketing bundle has an SEO pod that includes
seo-audit. It’s a good fit if you also want copy, CRO and content skills from the same source. - Install:
/plugin marketplace add alirezarezvani/claude-skillsthen/plugin install marketing-skills@claude-code-skills
5. seo-audit-skill (JeffLi1993)
- Source: github.com/JeffLi1993/seo-audit-skill · MIT · about 780 stars · last push 17 Jun 2026
- Free.
- What it does: a quick single-page audit, plus a separate
seo-audit-fullskill that adds Core Web Vitals and Open Graph checks. It hasn’t been updated since June, so check the issues tab before you depend on it. - Install:
npx skills add JeffLi1993/seo-audit-skill
6. seomachine (Craig Hewitt)
- Source: github.com/TheCraigHewitt/seomachine · MIT · about 7,600 stars · last push 23 Oct 2026
- Free repo. It requires a DataForSEO account, which is paid per use.
- What it does: a Claude Code workspace for content. It’s included here for its audit commands.
/analyze-existinggives a published post a 0 to 100 content health score with update priorities, and/optimizeruns a pre-publish SEO check. It’s a content audit tool, not a technical crawler. - Install:
git clone https://github.com/TheCraigHewitt/seomachine.git, thenpip install -r data_sources/requirements.txt, then runclaudeinside the folder.
7. claude-blog (AgriciDaniel)
- Source: github.com/AgriciDaniel/claude-blog · MIT · about 2,600 stars · last push 30 Oct 2026
- Free. The companion to claude-seo for writing and refreshing posts. Use it after the audit, on the content findings.
Directories worth browsing: ComposioHQ/awesome-claude-skills (about 78,500 stars, last push 18 Sep 2026) is a curated list, not an audit skill. claudemarketplaces.com and claudskills.com index dozens of community seo-audit variants. Read the SKILL.md before installing any of them, because quality varies a lot.
Paid data sources that make audit skills sharper
No standalone paid “SEO audit skill pack” turned up on Gumroad or the major directories during research. In practice, the paid side of this market is data: MCP servers and APIs that feed the free skills above.
8. DataForSEO MCP
- Source: dataforseo.com/pricing
- Paid, pay as you go. $50 minimum top-up, no subscription, balance doesn’t expire, $1 trial credit. The OnPage API is listed “from $0.00015 / page.”
- What it gives a skill: crawl data, SERPs, keywords and backlinks. claude-seo has a DataForSEO extension, and seomachine requires it.
9. Ahrefs MCP
- Source: docs.ahrefs.com/mcp
- Paid. Third-party reviews report that it needs the Lite plan or higher (Lite is widely cited at $129 a month) and that it uses your plan’s API units. Confirm on Ahrefs’ own pricing page before you budget for it.
- What it gives a skill: backlink, keyword and Site Audit data inside Claude.
10. SE Ranking MCP + Claude SEO Skills
- Source: seranking.com/claude-seo-skills.html
- Paid data, free skills. The skills install through the plugin marketplace. They need SE Ranking API credits (the page mentions free starter credits) to use live data. Ongoing pricing: check SE Ranking’s pricing page. This guide treats it as pricing not public for the MCP-only tier.
11. Firecrawl
- Source: firecrawl.dev/pricing
- Freemium. The free tier has 1,000 credits a month, and Hobby is $19 a month ($16 billed annually) for 5,000 credits. Firecrawl also publishes a
firecrawl-seo-auditworkflow skill (firecrawl/firecrawl-workflows, about 210 stars). - What it gives a skill: rendered page fetches at scale, which helps when Claude’s own web fetch is blocked or returns a stripped page.
Free vs paid comparison
| Option | Cost | Renders JS | Real CWV field data | Backlinks / rankings | Best for |
|---|---|---|---|---|---|
| claude-seo | Free (paid extensions optional) | Yes, headless Chromium | Only with Google APIs connected | Via paid extensions | Full technical + GEO audits |
| marketingskills | Free | No (uses Claude fetch) | No | No | Fast, editable on-page and schema checks |
| Anthropic Marketing seo-audit | Free | No | No | Only with MCP tool connected | Prioritised quick-wins plan |
| alirezarezvani marketing bundle | Free | No | No | No | Teams that want a wider skill library |
| JeffLi1993 seo-audit-full | Free | Limited | Lab only | No | Single-page spot checks |
| seomachine | Free repo + DataForSEO | No | No | Yes, via DataForSEO | Content refresh audits |
| DataForSEO MCP | $50 min top-up, per use | Via OnPage API | No | Yes | Cheap data at scale |
| Ahrefs MCP | Paid plan (reported Lite+) | n/a | No | Yes | Teams already on Ahrefs |
| SE Ranking MCP | API credits; pricing not public for MCP-only tier | n/a | No | Yes | Teams already on SE Ranking |
| Firecrawl | Free 1,000 credits / $19 mo | Yes | No | No | Fetching pages Claude can’t |
If you’re starting from zero, pick one free audit skill and add one paid data source only once you know which gap hurts. For a wider view of the SEO stack, see Claude Code skills for SEO.
How audit skills crawl or fetch pages
Short answer: most audit skills fetch one URL at a time with Claude’s web fetch tool, which returns a cleaned, markdown version of the page. The stronger ones also run a script (Python requests, a headless browser) to get the raw HTML, response headers and rendered DOM. Few of them do a true recursive crawl unless they call a crawler such as Firecrawl or DataForSEO OnPage.
That difference shapes what the audit can report:
- Cleaned fetch (web fetch tool). You get readable text and headings. You lose most of the
<head>: canonical tags, meta robots, hreflang and JSON-LD can be stripped or summarised. An instruction-only skill relying on this may “find” a missing canonical that’s actually there. - Raw HTTP fetch (script). You get status codes, redirect chains,
X-Robots-Tagheaders and the untouched source HTML. This is the right input for technical checks. - Rendered fetch (headless browser). You get the DOM after JavaScript runs. Comparing raw against rendered shows content, links or schema that only exist client-side.
- Crawler API. You get many URLs, internal link graphs, orphan pages and duplicate clusters. This is the only one of the four that approaches a Screaming Frog crawl.
Also check whether the skill reads robots.txt and sitemap.xml before choosing URLs. Without them, it samples whatever the homepage links to, and deep templates (filtered category pages, old blog archives) never get looked at.
A practical setup: let the skill audit the sitemap’s top templates (home, a category, a product or post, a location page) instead of trying to crawl everything. Templates repeat issues. One fix to a template can clear hundreds of URLs.
Takeaway: ask “raw, rendered, or cleaned?” for every finding about the <head>.
What each audit type should cover
Short answer: a useful audit covers five layers: technical (can it be crawled and indexed), on-page (does each page answer its query), schema (is the markup valid and truthful), Core Web Vitals (is it fast for real users), and GEO (can AI engines fetch, read and cite it). Here’s what a good skill checks at each layer, and the evidence you should expect it to show.
Technical audit
- Status codes, redirect chains and loops, soft 404s
robots.txtrules, meta robots,X-Robots-Tag- Canonical tags (raw vs rendered), duplicate URL patterns, parameters
- XML sitemap validity, and whether sitemap URLs return 200 and are indexable
- HTTPS, mixed content, hreflang reciprocity
- Internal linking depth and orphan pages (needs a crawl)
On-page audit
- Title and meta description length, uniqueness and keyword match
- One clear H1, a logical H2/H3 outline
- Thin or duplicate content, keyword cannibalisation between pages
- Image alt text, file size and modern formats
- Internal links with descriptive anchors
On-page is where Claude does well, because judging whether a page answers the query is a language task. For the keyword side, pair it with Claude Code skills for keyword research.
Schema audit
- JSON-LD present in raw HTML, not only injected by JavaScript
- Required properties for each type (Article, Product, FAQPage, LocalBusiness, Organization)
- Markup that matches visible content (a review rating that appears nowhere on the page is a policy risk)
sameAsand Organization entities for brand clarity
Confirm any schema verdict with Google’s Rich Results Test or the Schema.org validator. A skill can write valid-looking JSON-LD that still fails a required field.
Core Web Vitals audit
The current metrics are LCP (Largest Contentful Paint), INP (Interaction to Next Paint, which replaced FID in 2024) and CLS (Cumulative Layout Shift). Google ranks on field data from real Chrome users (CrUX), not lab tests. A skill without the PageSpeed Insights or CrUX API can only estimate from page weight, render-blocking resources and image sizes. That’s useful for finding causes, but it’s not proof that a page fails.
GEO audit
- AI crawler access in
robots.txt(GPTBot, ClaudeBot, PerplexityBot, Google-Extended) - Content in server-rendered HTML, since many AI crawlers don’t run JavaScript
- Answer-first passages, clear definitions, tables, dated facts
- An optional
llms.txtfile (a proposed convention, not a ranking standard) - Brand presence on sources AI engines cite often
claude-seo’s /seo geo and marketingskills’ ai-seo both cover this layer. For the writing side, see Claude Code skills for content writers.
| Layer | Claude does well | Needs outside data |
|---|---|---|
| Technical | Header, robots and canonical logic on fetched URLs | Site-wide crawl, log files |
| On-page | Intent match, outline, title quality | Search volume, rankings |
| Schema | Writing and spotting missing fields | Final validation (Google tools) |
| CWV | Diagnosing likely causes | Field data (CrUX / PSI API) |
| GEO | Citability and structure review | Actual citation tracking |
What audit skills can’t see
Short answer: a Claude Code audit skill can’t see Google’s view of your site. It has no Search Console index coverage, no real ranking history, no server logs, no full backlink graph, and no field Core Web Vitals unless you connect them. It also can’t see pages behind logins, bot walls or heavy client-side rendering that its fetch method doesn’t execute.
Spelled out:
| Blind spot | Why | How to close it |
|---|---|---|
| Index coverage and crawl stats | Only in Google Search Console | Export GSC reports, or connect a GSC MCP / claude-seo Google APIs |
| Real rankings and traffic | Not in the page | Ahrefs, SE Ranking or DataForSEO MCP; GA4 export |
| Field Core Web Vitals | CrUX data lives at Google | PageSpeed Insights API key |
| Backlink profile | Third-party index | Ahrefs / DataForSEO / SE Ranking |
| Server logs | On your server | Paste or upload log samples |
| Bot-blocked pages | Cloudflare or WAF challenges Claude’s fetcher | Firecrawl, or audit a staging copy |
| Interaction-gated content | Needs scroll or click | Headless browser scripts; claude-seo warns these can produce noisy results |
| Business context | Not on the page | Tell the skill which pages make money |
There’s a quieter risk too. Language models sound sure of themselves. A skill might report “no FAQ schema found” because its fetch stripped the <script> tag, not because the schema is missing. Ask it to quote the raw evidence (the header, the HTML snippet, the status code) for every high-severity finding. If it can’t, downgrade the finding to “needs verification.”
Takeaway: treat a Claude audit as a fast first pass and a hypothesis list. Google’s own tools are the final word.
How to turn audit findings into a prioritised fix list
Short answer: score each finding on impact, effort and confidence. Group findings by template, not by URL. Then ship in three waves: blockers (indexing and crawl), quick wins (titles, schema, internal links) and projects (speed, architecture, content). A 60-row list nobody fixes is worth less than ten tickets that ship.
Step 1: Deduplicate by template
Ask the skill to cluster findings by page type. “Missing meta description on 340 product pages” is one ticket for the product template, not 340 issues.
Step 2: Score impact, effort and confidence
| Score | Impact (1 to 3) | Effort (1 to 3) | Confidence (1 to 3) |
|---|---|---|---|
| 3 | Blocks indexing or affects money pages | Days of dev work | Verified with raw evidence or GSC |
| 2 | Affects rankings or CTR on many pages | A few hours | Plausible, partly verified |
| 1 | Cosmetic or best-practice | Minutes, CMS-level | Model inference only |
Priority = (Impact × Confidence) ÷ Effort. Sort descending.
Step 3: Ship in waves
- Blockers (this week): accidental
noindex, robots.txt disallows on key sections, broken canonicals, 5xx errors, redirect loops, sitemap listing non-200 URLs. - Quick wins (this sprint): title rewrites on high-impression pages, missing or invalid schema, internal links to orphaned money pages, AI crawler access.
- Projects (this quarter): LCP and INP work, JavaScript rendering fixes, information architecture, content consolidation.
Step 4: Write tickets a developer can act on
A prompt that works:
Turn these audit findings into tickets. For each: title, affected template,
evidence (quote the raw HTML/header), why it matters in one sentence,
exact fix, acceptance test, priority score. Mark anything without raw
evidence as "needs verification". Output as a markdown table.Step 5: Re-audit after fixes
Run the same skill on the same URLs after deploy and compare. claude-seo includes a /seo drift command for change tracking. With other skills, save the first report as a file and ask Claude to diff against it.
Common mistakes when auditing with Claude Code
Short answer: the usual mistakes are trusting cleaned fetches for <head> checks, treating lab Core Web Vitals as field data, auditing random URLs instead of templates, running a full multi-agent audit when a page check would do, and installing unvetted skills from directories.
- Believing every “missing” tag. Verify with view-source or
curl -I. - Mixing lab and field metrics. Report them separately.
- Running the heaviest audit by default. claude-seo’s README notes a full audit uses noticeably more of your Claude plan than a page check. Start with
/seo pageon one template. - Ignoring stale repos. Check the last commit date. SEO guidance changes (INP replaced FID, and FAQ rich results were restricted in 2023).
- Skipping the SKILL.md read. Skills can run scripts on your machine. Read what you install. If you’d like to tailor one, see how to build a Claude Code skill.
- No business context. Tell the skill which pages earn revenue, so priority reflects money rather than tag counts.
If you’re on a marketing team without a developer, Claude Code skills for non-developers and Claude Code skills for marketers cover the setup in plainer terms.
A starter audit workflow you can run today
- Install one free skill: claude-seo if you want depth, marketingskills if you want speed and readable instructions.
- Pull the sitemap and pick 4 to 6 template URLs.
- Run the page-level audit on each and save the output as
audit-v1.md. - Ask for raw evidence on every high-severity item, then spot-check three with
curl. - Add field data: a free PageSpeed Insights API key, plus a GSC coverage export.
- Generate the scored ticket table above.
- Ship wave one, re-run, and diff.
Add a paid data source (DataForSEO for low-cost per-use data, or the MCP of the tool you already pay for) only when rankings or backlinks become the bottleneck.
FAQ
Can Claude Code do an SEO audit? Yes. With an audit skill installed, Claude Code can fetch pages, check technical and on-page elements, review schema and write a prioritised report. It works best on a sample of templates, and it needs connected data for rankings, backlinks and field Core Web Vitals.
What is the best Claude Code skill for SEO audits?
For a free, full technical and GEO audit, claude-seo by AgriciDaniel is the most complete option found. For lighter checks you can read and edit, Corey Haines’ marketingskills seo-audit is a strong choice.
Are Claude Code SEO skills free? The skills listed here are free and open source (MIT or Apache-2.0). Costs come from your Claude plan and from optional data sources such as DataForSEO, Ahrefs, SE Ranking or Firecrawl.
How do I install an SEO audit skill in Claude Code?
Use npx skills add owner/repo, or inside Claude Code run /plugin marketplace add owner/repo then /plugin install. You can also copy a skill folder into ~/.claude/skills/.
Is there an official Anthropic SEO audit skill?
Yes. Anthropic’s knowledge-work-plugins repo includes a Marketing seo-audit skill that outputs quick wins and strategic fixes.
Can Claude Code crawl an entire website? Not by itself. Most skills fetch individual URLs. For full crawls, connect a crawler API (Firecrawl, DataForSEO OnPage) or use a skill with a crawl script, and expect to spend more of your plan.
Does Claude Code render JavaScript when auditing? Only if the skill runs a headless browser. claude-seo installs Chromium for this. Instruction-only skills rely on Claude’s web fetch, which doesn’t fully render client-side apps.
Can Claude Code measure Core Web Vitals? It can estimate likely causes from page weight and resources. For the field data Google uses, connect the PageSpeed Insights or CrUX API.
Can Claude Code replace Screaming Frog or Ahrefs? Not fully. It’s better at interpreting findings and writing fixes. It’s weaker at exhaustive crawling and has no backlink index. Many SEOs export crawler data and have Claude analyse it.
Does Claude Code connect to Google Search Console?
Yes, through an MCP server or skill integration. claude-seo has a /seo google setup for Google APIs. Without that, you can export GSC reports as CSV and give them to Claude.
What is a GEO audit? A Generative Engine Optimization audit checks whether AI engines such as ChatGPT, Perplexity, Gemini and Google AI Overviews can access, understand and cite your pages. It covers AI crawler rules, server-rendered content and answer-ready structure.
Should I allow GPTBot and ClaudeBot in robots.txt? If you want to be cited in AI answers, blocking them works against you. The decision depends on your content licensing stance, so treat it as a business choice and let the audit flag the current state.
Is it safe to install SEO skills from directories? Read the SKILL.md and any scripts first. Skills can execute code locally. Prefer repos with an open license, recent commits and visible maintainers.
How much does a Claude Code SEO audit cost? The skill itself is usually free. You pay through your Claude subscription or API usage, plus any data source. DataForSEO, for example, starts with a $50 top-up and per-request pricing.
Can Claude fix the SEO issues it finds? In a codebase, yes. Claude Code can edit templates, add JSON-LD, rewrite titles and update robots.txt. Review each change and re-run the audit afterwards.
Sources and methodology
- GitHub API (stars, license, last push), queried 1 Nov 2026, for every repo listed
- AgriciDaniel/claude-seo README
- coreyhaines31/marketingskills README
- anthropics/knowledge-work-plugins seo-audit listing
- alirezarezvani/claude-skills README
- TheCraigHewitt/seomachine README
- DataForSEO pricing · Firecrawl pricing
- Ahrefs MCP docs · SE Ranking Claude SEO Skills
FAQ questions came from People Also Ask phrasing and skill-directory questions. Targeted Reddit searches returned no indexed threads specific to Claude Code SEO audits at the time of research. No hands-on benchmark of the skills was run for this guide, and the comparison reflects documented capabilities.
Your next step
Pick one template URL, install one free skill, and run a single page audit today. Then make it ask for raw evidence before you believe anything it says. For everything else in this series, start at Claude Code skills for SEO, or see how the same approach applies to copy in Claude Code skills for copywriters.
