Technical SEO with Claude + Semrush MCP: Step-by-Step Site Audit Triage and Fix Prioritization

• Every recommendation cites issue IDs and example URLs.
• If the mechanism isn’t clear, label it needs investigation and request the next artifact.
• Keep scope to triage + prioritization, not a technical SEO primer.

Step 1 — Pull structured Semrush Site Audit data via MCP (ticket-oriented export)

Your scoring and tickets are only as good as the export. Pull a minimal, stable schema you can merge across batches:

• issue_id (stable key for dedupe)
• issue_name, issue_type, semrush_severity
• description
• affected_urls_count
• examples (2–5 URLs)
• template_or_pattern_hints (paths, params, page type hints)
• notes_or_recommendations
• project, crawl_timestamp

Export in batches (by category or severity), but keep identical keys across chunks. Preserve issue_id so Claude can ignore duplicates across imports.

Data hygiene that prevents bad prioritization: normalize near-synonyms (“Duplicate title tags” vs “Duplicate titles”), dedupe repeats across subdomains/protocols, and consolidate http:// + https:// variants when it’s clearly one remediation.

Mini-example: one issue record (JSON-like)

{
  "issue_id": "SA-20491",
  "issue_name": "5xx server errors",
  "issue_type": "Crawlability",
  "semrush_severity": "High",
  "description": "Pages return 5xx status codes during crawl.",
  "affected_urls_count": 37,
  "examples": [
    "https://example.com/category/widgets",
    "https://example.com/product/blue-widget"
  ],
  "template_or_pattern_hints": ["/product/", "category template"],
  "notes_or_recommendations": "Investigate server logs and fix errors; validate response codes.",
  "crawl_timestamp": "2026-10-18T10:20:00Z"
}

Step 2 — Classify issues into engineering themes (triage taxonomy)

Classify each issue two ways: (1) crawlability vs indexability, and (2) template-level vs URL-level. This cuts noise and prevents “loud” findings from becoming the wrong work.

Crawlability means bots can’t reliably fetch pages: 4xx/5xx, redirect loops, blocked resources, timeouts, recurring fetch failures.

Indexability means bots can fetch pages, but signals prevent indexing or consolidate to another URL: noindex, canonical to a different URL, meta robots directives, soft-404 behavior.

Template vs URL level: high affected_urls_count plus repeating patterns (/product/, ?sort=) or shared directives usually points to a template/system cause. Low counts without a pattern are often URL-specific.

Decision rule: mark ready to fix only when audit evidence points to a clear mechanism. Otherwise mark needs investigation and ask for one next input (headers, HTML snippet, logs, redirect chain).

Step 3 — Prioritize with an impact/effort/risk rubric (the scoring model)

Semrush severity is a signal, not your backlog order. Let it inform Impact and Risk, then let Scope and Effort do the sorting.

Score each issue 1–5:

• Impact (organic + user): restores crawling, prevents de-indexing, consolidates signals.
• Scope (URLs/templates): anchor to affected_urls_count; template bugs usually outrank single-URL problems.
• Effort (dev complexity): simple config vs template work + deploy + QA.
• Risk (regression/SEO risk): chance of accidental de-indexing, canonical breakage, redirect chains.

Formula:

Priority = (Impact × Scope) − (Effort + Risk)

Rule: severity informs Impact/Risk but doesn’t override Scope/Effort. A “High” item affecting 3 URLs can land below a “Medium” item affecting 20,000.

Output: a ranked list with Top 10 “do next” plus a Later backlog, each with a one-line rationale (“moved up: template-level + high scope” / “moved down: high regression risk”).

Step 4 — Turn top findings into ticket-ready fixes (Claude prompt + ticket template)

Use one ticket format so engineering doesn’t have to translate SEO language.

Ticket template:

• Problem statement
• Evidence (Semrush issue_id, severity, counts, 2–5 example URLs)
• Root cause hypothesis (label as hypothesis)
• Recommended fix
• Alternatives (1–2)
• Risks/side effects
• Acceptance criteria (measurable “done”)
• Validation plan

Claude prompt pattern (evidence-led):

Use ONLY the evidence provided (issue_id, description, example URLs) and quote the fields you rely on.

Output: (1) recommendation (2) alternatives (3) risks (4) test cases.

If evidence is insufficient, label “needs investigation” and ask for the next specific artifact.

Acceptance criteria examples: status codes/redirects (no loops; chain length ≤ 1), canonicals resolve 200 and match intent, titles/descriptions unique on the affected template (20-URL spot-check), HTTPS links resolve cleanly, and noindex patterns changed only after intent review. In my experience, splitting “template fix” from “cleanup/backfill existing URLs” avoids partial releases.

Step 5 — Validate and close the loop (re-crawl + targeted checks)

After shipping the top 1–3 fixes, re-run Semrush Site Audit and compare deltas. Expect reduced affected URL counts and/or lowered severity for the same issue IDs. Some items won’t move if you intentionally deferred them.

Spot-check the exact URLs cited in tickets with Seosoft tools:

• Meta Tags Checker (title/description/canonical): https://seosoft.com/tools/meta-tags-checker
• SSL Checker (HTTPS/certs/redirect behavior): https://seosoft.com/tools/ssl-checker

If you’re diagnosing bot-specific behavior, use User Agent Parser only to verify the UA string you’re seeing: https://seosoft.com/tools/user-agent-parser

Keep a small regression checklist (3–7 items) tied to what you shipped: no unintended noindex on key templates, no new redirect loops, canonicals resolve 200, HTTPS enforced, and no fresh 5xx spikes.

Conclusion

You should now have a ranked backlog (top 10 + later) and ticket-ready fixes tied to specific Semrush issue IDs and example URLs. Keep Claude on evidence: if it can’t point to the audit row, treat it as investigation, not a fix. Implement the top 1–3, re-crawl, spot-check the ticket URLs with Meta Tags Checker and SSL Checker, then carry your regression checklist into the release process so the same issues don’t return.

Sources

1. Semrush Site Audit