A Technical SEO Audit Claude Skill That Runs on Your Own Crawl Data

Most SEO tools ask for a subscription before they'll crawl a single page. The technical SEO audit Claude skill built by developer Ian Nuttall skips that trade entirely: it's a local CLI and MCP server, published as the npm package seo, that runs against a site's own crawl data plus first-party Search Console and Google Analytics access. No dashboard login, no third-party crawl budget, no mandatory API key. As of today, the repo sits at 206 stars and 16 forks on GitHub, Apache-2.0 licensed.

What it actually does

Install with npm i -g seo, then run seo start to walk through Google sign-in for Search Console and an optional Google Analytics or Clicky connection, or skip sign-in entirely and run seo report --url https://example.com for a pure technical crawl with no account needed. The tool then produces a ranked list of findings, not just a wall of warnings; it weighs each issue by affected URL count, rule severity, and actual search visibility from Search Console data, so a broken canonical on a page with real traffic outranks the same issue on an orphaned page nobody visits.

It ships as a Claude- and Cursor-compatible skill (seo skill list) and installs into MCP clients directly (seo mcp install), so an agent can call seo_list_reports, seo_describe_report, and seo_run_report against the same report catalog the CLI uses, and get back deterministic JSON instead of screen-scraped dashboard text.

How the local crawl actually works

The crawler runs two distinct passes. A health-check mode reads the sitemap and validates each URL without rendering the page body: HTTP status, redirect chains, robots.txt directives, and access blocks, scaling request concurrency only after the first requests succeed cleanly. A full crawl mode goes deeper, pulling metadata, canonicals, link structure, and structured data, bounded by --max-pages, --max-depth, --include, and --exclude flags so a large site doesn't run away with itself.

The audits themselves cover specific, checkable things: title and meta description quality, internal and external link validation, indexability signals (robots.txt, noindex tags, canonical declarations), structured data markup, Core Web Vitals and performance, HTTPS and security headers, mobile responsiveness, hreflang and international SEO setup, and social preview completeness (Open Graph, Twitter cards). Past the base 70-plus checks, the report catalog extends into specific workflows: quick-wins for pages ranking 4 to 10 with low click-through, second-page for URLs stuck at positions 10 to 20, striking-distance, decaying-pages, cannibalisation, and a technical-watch mode built to catch regressions after a deploy.

What's actually free versus what costs money

The core loop, crawl plus report plus Search Console and Google Analytics reads, is free and stays local: tokens, project profiles, and cached reports live on the machine, not a vendor's server. That matches the "uses your own crawl data" pitch. Where cost enters is optional: keyword and competitor research through providers like SerpBase, DataForSEO, Semrush, or Ahrefs, connected separately with their own credentials and their own per-request pricing. A site owner who only wants a technical audit never needs to touch that layer. An agency running competitor keyword-gap analysis will.

One design detail worth noting: the tool is explicit that partial or capped crawl data is never reported as a clean zero, and heuristics are labeled as heuristics rather than presented as confirmed ranking factors. That's a deliberate anti-hallucination stance for a tool meant to hand its output straight to an agent.

Where this overlaps with other picks here, and where it doesn't

AIOS Guide has already covered ai-for-seo, a general small-business SEO primer, and the ai-search-console plus geolook pairing for tracking whether a brand gets cited inside AI answers like ChatGPT or Google's AI Overviews. This skill is neither of those. It's a technical and on-page audit tool: crawl-level findings about metadata, links, indexability, and structured data, backed by real Search Console numbers rather than general advice or citation tracking. It does have a few adjacent AI-search reports (ai-readiness, geo-gaps, ai-mention-research), but they sit as a handful of report types inside a much larger technical-SEO catalog, not the tool's reason for existing.

Who should install it

Marketers and SEOs running their own sites or client audits get the clearest win: a real crawl instead of a manual checklist, with findings ranked by actual visibility data rather than generic rule counts. Developers get a CLI they can drop into a CI pipeline to catch SEO regressions before a release ships. The rough edge is Node 22 as a hard requirement, and the fact that the deepest competitive research (rank tracking, domain overviews, keyword gaps against named competitors) needs a paid provider connection on top of the free core. For a straight technical audit of one site, though, the free path covers the whole job.

Repo: github.com/iannuttall/seo