/seo-site-audit

name: seo-site-audit
description: Run a complete technical SEO + AI-search audit of a website with crawlie.
  Use when the user asks to audit, crawl, or check a site for SEO issues, broken
  links, redirects, missing titles/descriptions/metadata, canonical or heading
  problems, image alt text, duplicate or thin content, or generative-engine (GEO /
  AI-search) readiness. Works even if crawlie is not installed.

SEO Site Audit (powered by crawlie)

Deliver a prioritized, plain-English technical SEO + GEO audit of a site. crawlie is a fast OSS crawler that ships on npm as `crawlie` — you do **not** need it pre-installed and you do **not** need its source repo.

Bootstrap — make crawlie runnable

Decide once how you'll invoke crawlie, then reuse that prefix for every command:

1. If `mcp__crawlie__*` tools exist this session, **prefer them** (`crawl_site`, `audit_url`, `explain_issue`, etc.) and skip the CLI entirely. 2. Else pick a CLI prefix:

• If `crawlie` is on PATH (`command -v crawlie`) → use `crawlie`.
• Otherwise → `npx -y -p crawlie crawlie`

(downloads + caches from npm on first use; needs only Node/npx — this *is* the install). 3. If there's neither the MCP nor Node, stop and tell the user: "Install Node (https://nodejs.org) and I can run crawlie with no further setup, or install the crawlie MCP for first-class tools."

Below, `<crawlie>` means the prefix you chose in step 2 (or the equivalent MCP tool).

Workflow

1. **Scope.** Confirm the start URL. Whole site, or a section? For a section add `--include '<glob>'` (repeatable); skip areas with `--exclude '<glob>'`. 2. **Crawl** as machine-readable JSON:

   <crawlie> crawl <url> --format json

• Large sites: cap with `--max-pages N` (default 500) and/or `--max-depth N`.
• Tune speed with `-c <concurrency>` (default 16) and `--timeout <secs>`.
• MCP equivalent: `crawl_site` with the same options.

3. **Triage** the findings: group by severity (error → warning → notice), then by `rule` id. Count affected URLs per rule. Lead with the highest-impact errors. 4. **Explain** the top 3–5 issue types so fixes are concrete:

   <crawlie> explain <rule-id>

(MCP: `explain_issue`.) Real rule ids you'll see include: `broken-link`, `redirect-chain`, `client-error`, `server-error`, `title-missing`, `title-too-long`, `title-duplicate`, `description-missing`, `description-duplicate`, `h1-missing`, `h1-multiple`, `canonical-missing`, `image-missing-alt`, `lang-missing`, `viewport-missing`, `og-missing`, `twitter-missing`, `mixed-content`, `not-secure`, `no-hsts`, `duplicate-content`, `thin-content`, `low-text-ratio`, `blocked-by-robots`, `hreflang-incomplete`, and the GEO set: `geo-not-answerable`, `geo-no-structured-data`, `geo-no-llms-txt`, `geo-no-semantic-html`, `geo-no-author`, `geo-thin-for-ai`. 5. **Deliver** a report, not raw JSON:

• Headline: the **Health** score (technical SEO) and **GEO** score (AI-search

readiness) crawlie returns.

• A prioritized table: rule · severity · affected-URL count · one-line fix.
• The top fixes spelled out with example URLs.
• Offer a shareable HTML report: `<crawlie> crawl <url> --format html -o report.html`.

Tips

• Audit just a few specific pages instead of crawling: `<crawlie> audit <url> [<url> ...]`

(MCP: `audit_urls`).

• Save to local history for later diffing: add `--save` (then `<crawlie> reports`,

`<crawlie> report <id>`).

• Never dump the full JSON at the user — summarize, prioritize by impact, link the report.

Make it permanent (optional — offer once, don't block)

If you ran via npx, you may offer: "Want crawlie installed for good? `npm i -g crawlie`. To get first-class agent tools, also run `claude mcp add crawlie crawlie-mcp` — those activate after you restart Claude."