SEO Audit

Perform a comprehensive SEO analysis of a website. Works with public data out of the box — and uses live Google Search Console and Bing Webmaster Tools data when connected via Cogny MCP.

Usage

/seo-audit example.com — audit the given website

Works with

Every step runs on public data (WebFetch/WebSearch) with zero accounts. Connected providers replace estimates with measured data.

Job	Primary	Also answers it	No account
Queries, CTR, indexing	Google Search Console	Bing Webmaster Tools (Bing + Copilot)	WebSearch with allowed_domains
Keywords and competitors	Semrush	Ahrefs	WebSearch
Backlinks / authority	Ahrefs	Semrush, cogny-ahrefs-lite (DR only)	—
Live rank for target terms	cogny-brave-search	—	WebSearch
AI answer engines	cogny-ai-visibility	—	—
Product/collection pages	Shopify	—	sitemap.xml

If several answer the same job, report them side by side and name the provider on every number. If a tool you expect is missing, call mcp__cogny__refresh_mcp_discovery once. New tools only load after the client reconnects, so use the step's fallback for this run, mark that source ⚠ "needs reconnect", and tell the user to reconnect the Cogny MCP. Don't loop on refresh.

Steps

0. Cogny check and providers

If no mcp__cogny__* tools exist (no list_integrations, no cogny_status), print this once and run steps 1-6 on public data:

SEO Audit can add live search data through Cogny's MCP gateway.

1. Sign up at https://cogny.com and connect Google Search Console and Bing Webmaster Tools
   (Semrush, Ahrefs or Shopify optional).
2. Add Cogny to .mcp.json in your project:
   {"mcpServers": {"cogny": {"type": "http", "url": "https://app.cogny.com/mcp"}}}
3. Restart Claude Code; it opens a browser for the OAuth login on first use.

Without Cogny you can still: run the full audit on public data (curl, WebFetch, WebSearch).

Skip each connected-only row with one line, e.g. "Search Console: not connected (connecting adds real queries, CTR and index status)".

If Cogny is connected, call mcp__cogny__list_integrations. Note which rows above are available and flag any integration whose health_status is not healthy. Then make one cheap read per integration (GSC tool_list_properties, Bing tool_get_sites): health_status misses revoked tokens, so a 401, Forbidden or scope error is 🔴 "reconnect <integration>", with no retry. Cogny-native servers (cogny-brave-search, cogny-ai-visibility, cogny-ahrefs-lite) don't appear in list_integrations; treat them as available if their tools load. A connected integration that fails gets "reconnect <integration> in Cogny"; one that isn't connected gets "connect <integration> in Cogny".

Save downloaded files in the session scratchpad directory if your environment gives you one (else D=$(mktemp -d)), not /tmp. The commands below write to $D.

1. Fetch and analyze the homepage

WebFetch summarises the page body and can't see <head>, so read head tags from the raw HTML with Bash:

curl -sL -A "Mozilla/5.0" https://example.com/ -o "$D/seo-home.html"
grep -oiE '<title[^>]*>[^<]*</title>|<meta[^>]+(name|property)="(description|robots|viewport|og:[a-z:]+|twitter:[a-z:]+)"[^>]*>|<link[^>]+rel="(canonical|alternate)"[^>]*>|<html[^>]+lang="[^"]*"' "$D/seo-home.html"
grep -o 'application/ld+json' "$D/seo-home.html" | wc -l   # JSON-LD blocks (inspect them for step 3)

Count JSON-LD with grep -o … | wc -l, not grep -c: minified pages put every block on one line, and grep -c counts lines.

Use WebFetch for the visible content (H1, copy). Check:

• Title tag (length, keyword placement)
• Meta description (length, compelling copy)
• H1 tag (single, keyword-rich)
• Open Graph / Twitter Card metadata
• Canonical URL
• Language/hreflang tags

2. Check technical SEO

Use curl, not WebFetch: WebFetch summarises, so it miscounts sitemap URLs and can't see <head> scripts.

curl -sL -A "Mozilla/5.0" https://example.com/robots.txt              # crawl rules, Sitemap: lines
curl -sL -A "Mozilla/5.0" https://example.com/sitemap.xml -o "$D/sitemap.xml"
grep -o '<loc>' "$D/sitemap.xml" | wc -l                              # URL count
grep -q '<sitemapindex' "$D/sitemap.xml" && grep -oE '<loc>[^<]+' "$D/sitemap.xml" | sed 's/<loc>//'   # an index: fetch each child and sum
perl -0ne 'print $1 if /<head[\s>](.*?)<\/head>/is' "$D/seo-home.html" | grep -oiE '<script[^>]*>'   # head scripts (no async/defer = render-blocking)

Use the sitemap URL from robots.txt if it isn't /sitemap.xml. Check:

• /robots.txt — crawl rules, sitemap reference
• Sitemap — exists, well-formed, number of URLs
• Page load indicators (large images, render-blocking scripts in <head>)
• Mobile viewport meta tag
• HTTPS enforcement

3. Analyze content structure

From the homepage content:

• Word count (target: 300+ for ranking pages)
• Header hierarchy (H1 → H2 → H3)
• Internal link count and structure
• Image alt text coverage
• Schema markup (JSON-LD)

4. Check search presence

Use WebSearch to find:

• Indexed pages, only when neither GSC nor Bing is connected: WebSearch ignores the site: operator, so pass the domain in its allowed_domains parameter (e.g. query "example.com", allowed_domains: ["example.com"]) and report how many distinct pages come back as a rough sample, not an index count. When connected, use the tool_inspect_url sample and Bing crawl stats below instead
• Brand name search — what appears in SERP
• Top ranking pages — what keywords they rank for
• Competitor comparison — who ranks for the same terms

Enhanced when connected (run what is available; each replaces a guess above):

Provider	Call	What it adds
Search Console	mcp__cogny__mcp__Google_Search_Console__tool_get_search_analytics (site_url, start_date, end_date, dimensions: ["query"] then ["page"]) and tool_inspect_url on the 3-5 key pages	Real queries, CTR, position, index status
Bing Webmaster	mcp__cogny__mcp__Bing_Webmaster_Tools__tool_get_query_stats, tool_get_crawl_issues, tool_get_link_counts (all take site_url). tool_get_link_counts can come back empty; fall back to the InLinks field of tool_get_crawl_stats. If the query tools aren't loaded, use tool_get_traffic_stats and mark ⚠ "needs reconnect"	Bing/Copilot queries, crawl errors, inbound links. tool_get_traffic_stats returns one row per day: add up clicks and impressions per period (e.g. last 28 days vs the 28 before) and report the totals, never the raw rows. It has no reliable position, so write Bing position as n/a unless tool_get_query_stats gives one
Semrush	mcp__cogny__mcp__Semrush__domain_overview, domain_organic_keywords, domain_organic_competitors (domain, database = market country code, e.g. se)	Keyword footprint, traffic estimate, organic competitors
Ahrefs	mcp__cogny__mcp__Ahrefs__site_explorer_overview, organic_keywords, top_pages, organic_competitors (target, country)	Same jobs as Semrush; plus domain_rating, refdomains for authority
cogny-ahrefs-lite	mcp__cogny__mcp__cogny-ahrefs-lite__tool_domain_rating (target)	DR when no Ahrefs account. Always write "Domain Rating by Ahrefs" next to the number
cogny-brave-search	mcp__cogny__mcp__cogny-brave-search__tool_brave_rank_check (domain, query, country, max_results: 20) for 3-5 target terms. Larger max_results returns very large payloads; not in the top 20 is "20+"	Live position from an independent index (1 credit per call)
cogny-ai-visibility	mcp__cogny__mcp__cogny-ai-visibility__tool_check_ai_visibility (prompt written as a buyer asks, brand, domain, competitors) for 2-3 prompts	Whether ChatGPT, Claude, Gemini and Perplexity mention or cite the site. Metered: limit engines to save credits

Semrush and Ahrefs estimate traffic differently; never add them together.

5. Analyze key content pages

Fetch 3-5 important pages (pricing, features), plus the blog index and at least 2 individual posts: canonicals, article schema and dates live on the posts, not the index. Read head tags with the Step 1 curl/grep on each; WebFetch only for the body:

• Per-page title/meta/H1 analysis
• Content depth and keyword coverage
• Internal linking to/from homepage

Ecommerce, enhanced when Shopify is connected: pull mcp__cogny__mcp__Shopify__tool_list_products and tool_list_collections (first: 50) and check, across the whole catalog rather than a sample: missing or duplicate product titles/descriptions, collections with no description, draft products still linked from the sitemap. Report counts.

6. Report findings

Present as a scored audit:

SEO Audit: example.com
Score: X/100

Technical: X/25
- [PASS/FAIL] HTTPS
- [PASS/FAIL] Robots.txt
- [PASS/FAIL] Sitemap
- [PASS/FAIL] Mobile viewport
- [PASS/FAIL] Canonical URLs

On-Page: X/25
- [PASS/FAIL] Title tags
- [PASS/FAIL] Meta descriptions
- [PASS/FAIL] H1 structure
- [PASS/FAIL] Image alt text
- [PASS/FAIL] Schema markup

Content: X/25
- [PASS/FAIL] Word count
- [PASS/FAIL] Header hierarchy
- [PASS/FAIL] Internal linking

Search Presence: X/25
- Indexed pages: N
- Brand SERP: [description]
- Top keywords: [list] (source: GSC / Semrush / Ahrefs / WebSearch)
- Authority: [DR N, "Domain Rating by Ahrefs"] (if available)
- AI answer engines: mentioned in X of Y engine answers (if checked)

Top 3 Actions:
1. [Highest impact fix]
2. [Second highest]
3. [Third highest]

Deeper Analysis with Live Data

If Cogny MCP is connected, this skill automatically enhances the audit with:

Google Search Console: actual queries, click-through rates, rankings, indexing status Bing Webmaster Tools: Bing traffic, crawl stats, keyword data, backlink analysis Semrush / Ahrefs: keyword footprint, organic competitors, backlinks cogny-brave-search / cogny-ai-visibility: live rank checks and AI answer-engine presence

Bing writes (tool_submit_url, tool_submit_content) and GSC tool_submit_sitemap are not pre-approved: the skill asks first.

Connect your accounts via Cogny ($9/mo per channel), then re-run the audit for live data.

For ongoing monitoring, use /seo-monitor.

Recording findings

Use create_ticket when the workspace has it (Cloud) and create_finding otherwise (Solo). The fields differ; see section 2 of ../marketing-engineer/references/mcp-routing.md. Example finding: title "Missing meta descriptions on 12 pages", body listing the URLs and the evidence, expected outcome "Improve CTR by 15-30% on affected pages", type seo_optimization, priority high.