Technical SEO audit toolkit for Claude Code. Multi-agent architecture with specialized skills — each audit runs like a consulting team: a maieutic project lead, technical specialists, a senior editor, and a QA reviewer.
Built by Kōeki Agency for SEO consultants who want repeatable, evidence-based audits with a clear audit trail.
Most SEO audit tools crawl your site and dump 300 issues color-coded. They don't know your context, they can't compare an old menu to a new one before deployment, and they produce reports nobody reads.
This toolkit runs audits as a collaboration between specialized agents, coordinated by an orchestrator that actually talks to you before doing anything. Each specialist has narrow expertise (semantic HTML, link equity, crawlability, accessibility, performance, information architecture), their findings are consolidated, a senior editor writes the report, and a QA reviewer challenges it before delivery.
Every claim in the final report is tagged: I KNOW (verified in the data), I THINK (interpretation from a documented source), or I CANNOT VERIFY (needs live access). That's what separates a serious audit from a bullshit one.
No live site access required. The fetcher handles public URLs via curl, JS-heavy SPAs and authenticated staging environments via Playwright MCP.
git clone https://github.com/agencekoeki/seo-audit-for-claude-code.git
cd seo-audit-for-claude-code
claudeThen in Claude Code, type:
/audit-menu
The orchestrator will interview you about what you need (single audit vs before/after comparison, URLs, authentication, specific pages to check) and coordinate the audit end-to-end.
The audit runs as 10 specialized roles working in coordination:
Coordination
orchestrator— Maieutic project lead. Interviews you, dispatches work, consolidates findings.reporter— Senior editor. Turns technical findings into a readable client report.reviewer— QA / devil's advocate. Challenges every verdict before delivery.
Technical specialists
fetcher— Data acquisition. Handles public URLs, authenticated pages, JS-rendered SPAs.semantic— HTML5 & ARIA expert.<nav>,<ul>/<li>, accessibility attributes.link-equity— Internal linking analyst. Google patents, PageRank distribution, dilution.crawlability— Crawl & rendering specialist. HTML source vs DOM after JS.accessibility— Mobile-first & WCAG 2.1/2.2.performance— Core Web Vitals impact of navigation (INP, CLS, LCP).architecture— Information architecture, user journey, content silos.
Each specialist has its own context, its own tool permissions, and its own knowledge base. They communicate via structured JSON findings.
Audit mode — Diagnose an existing menu. Each specialist analyzes their dimension. Consolidated report with verdicts.
Comparison mode — Validate a menu redesign against the current one. The comparator flags: removed URLs (orphan risk), depth changes, anchor text shifts, semantic regressions. Critical for pre-deployment gating.
Local-first. No SaaS, no tracking. Python 3.10+ standard library. Evidence-based. Every rule traces to a primary source: Google patents (Reasonable Surfer US 7,716,225, Boilerplate Detection US 8,898,296), Web.dev CWV documentation, WCAG specs. Maieutic over mechanical. The orchestrator talks to you. It asks, reformulates, confirms. Honest about limits. Every report lists what was checked, interpreted, or couldn't be verified.
- Claude Code ≥ 2.x (subagents support required)
- Python 3.10+
- Playwright MCP (optional — for authenticated/JS-heavy sites)
seo-audit-for-claude-code/
├── CLAUDE.md # Instructions for Claude Code
├── .claude/
│ ├── agents/ # The 10 specialized roles
│ ├── skills/ # Reusable expertise with scripts + references
│ └── commands/
│ └── audit-menu.md # /audit-menu entry point
├── shared/ # Shared Python modules
├── knowledge/ # Global SEO knowledge base
├── audits/ # Outputs (one folder per audit, gitignored)
└── tests/
See CLAUDE.md for the detailed architecture.
MIT
Sébastien Grillot — Kōeki Agency. SEO since 2008.