Best SEO skills for Claude Code: audits, GEO, programmatic and keyword research

Picking an SEO skill for Claude Code comes down to the job: audit, technical fixes, AI search, programmatic pages, or keyword research. Each has a clear leader in the catalog, and the ranked list of marketing and SEO skills shows the rest.

Key takeaways

• The seo-audit skill leads with 240,705 installs (skills.sh registry) from a repo with 47,132 stars (GitHub).
• For AI search, ai-seo has 153,490 installs and seo-geo has 54,505 installs (skills.sh registry).
• For programmatic pages, programmatic-seo has 152,370 installs (skills.sh registry).
• firecrawl-seo-audit is the only skill here that declares a required environment key, FIRECRAWL_API_KEY.
• The three coreyhaines31/marketingskills skills share one repo, so the 47,132 stars (GitHub) belong to the repo, not to each skill.

Which SEO skill should you install first in Claude Code?

Install seo-audit first. It has 240,705 installs (skills.sh registry) and works on any site without keys: the skill tells Claude to review five areas in a fixed order (crawlability and indexation, technical foundations, on-page, content quality, authority) and to return an executive summary, impact-rated findings with evidence, and a prioritized fix roadmap.

npx skills add https://github.com/coreyhaines31/marketingskills --skill seo-audit

One detail matters for anyone running WordPress. The skill warns that web_fetch and curl cannot reliably see structured data, because plugins such as Yoast, AIOSEO and RankMath can inject JSON-LD with client-side JavaScript. It tells Claude to render the page in a browser or use Google's Rich Results Test before it reports "no schema found". That one rule prevents a common false finding. It also lists the Core Web Vitals targets it checks: LCP under 2.5s, INP under 200ms, CLS under 0.1.

Is there a full Claude SEO suite with subagents?

Yes. claude-seo (skill name seo) has 9,287 installs (skills.sh registry) and 16,391 stars (GitHub), the second most stars of any audit skill here, after the marketingskills repo (47,132 stars). Its SKILL.md describes 25 sub-skills and 19 sub-agents. A /seo audit <url> run detects the business type (SaaS, local, ecommerce, publisher, agency) and spawns subagents such as seo-technical, seo-content, seo-schema, seo-sitemap, seo-performance, seo-visual and seo-geo.

Other commands include /seo page, /seo schema, /seo cluster <seed-keyword>, /seo hreflang, /seo backlinks and /seo drift baseline <url>, which captures a baseline you can compare after a deploy. It runs bundled Python tools through a launcher, so expect a setup step (/seo setup).

npx skills add https://github.com/AgriciDaniel/claude-seo --skill seo

If you want technical checks tied to Lighthouse, addyosmani/web-quality-skills seo has 52,706 installs and 2,754 stars. It runs Lighthouse SEO and Agentic Browsing checks and inspects response headers, redirects, robots.txt and sitemap coverage. It separates crawl findings from content quality and does not promise ranking changes.

Which skill handles AI search (GEO and AEO)?

Use ai-seo for strategy and seo-geo for a hands-on checklist. ai-seo (153,490 installs, skills.sh registry) targets citations in ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews. It routes to reference files on agent readiness, answer accuracy, prompt portfolios, platform ranking factors and llms.txt.

seo-geo (54,505 installs, 1,762 stars) bundles the nine Princeton GEO methods in a table, lists the crawlers robots.txt should allow (Googlebot, Bingbot, PerplexityBot, ChatGPT-User, ClaudeBot, GPTBot), and ships a free scripts/seo_audit.py that checks title, meta, H1, robots, sitemap and load time without an API.

npx skills add https://github.com/ReScienceLab/opc-skills --skill seo-geo

How do you do programmatic SEO with Claude Code?

Install programmatic-seo, which has 152,370 installs (skills.sh registry). It defines 12 playbooks (templates, curation, conversions, comparisons, locations, integrations, glossary, directory and others) and ranks data by defensibility, from proprietary down to public sources. It prefers subfolders over subdomains and hub-and-spoke internal links, and it says 100 great pages beat 10,000 thin ones. Google's Search Central documentation is the reference to check when your pages look like doorway pages.

What is the best keyword research skill and MCP?

The OpenSEO keyword-research skill has 6,935 installs (skills.sh registry) and the repo has 17,293 stars (GitHub). It needs the OpenSEO MCP server and a projectId. The workflow starts from your own Search Console data: it pulls queries at positions 5 to 20 with at least 50 impressions, then calls get_keyword_metrics (up to 700 keywords per call) for difficulty and intent. research_keywords takes 1 to 5 seeds per call, and save_keywords runs only after you confirm.

npx skills add https://github.com/every-app/open-seo --skill keyword-research

A similar option is iannuttall/seo (15,039 installs, 468 stars). It reads Search Console, GA4, Bing Webmaster Tools and Clicky through a CLI and MCP server. seo crawl --url <url> needs no sign-in, and keyword data comes from DataForSEO or Ahrefs. Finally, firecrawl-seo-audit (32,108 installs, 151 stars) scrapes pages with Firecrawl and refuses to invent a score or traffic forecast.

A complete starter stack

1. seo-audit - the baseline audit and fix roadmap.
2. addyosmani/web-quality-skills seo - Lighthouse-backed technical checks after each change.
3. ai-seo - citation-ready content and agent readiness.
4. keyword-research (OpenSEO) - striking-distance queries from Search Console.
5. programmatic-seo - only once you have a data source worth a template.
6. claude-seo - add it when you want one command to fan out to subagents.

Install notes for Claude Code are in the official skills documentation.

Common pitfalls

• Trusting a "no schema" finding from curl output. JS-injected JSON-LD is invisible to it.
• Running programmatic-seo on public data alone. The skill ranks that as the weakest data source.
• Installing every skill at once. Their descriptions overlap, so Claude may pick the wrong one. Start with two and add a third when a gap shows.
• Treating audit output as a ranking promise. Several of these skills say outright not to forecast traffic.

Pick one audit skill, fix what it finds, then add a specialist for the gap that remains.

Common questions