atomno-mcp/mcp-seo-audit

MCP (Model Context Protocol) server for **technical SEO & GEO audits**, powered by
the detail.web engine — **real measurements, not LLM
guesses**. Run it from Cursor, Claude Desktop or any MCP client. **8 tools:**
audit_site (deterministic 0–100 health score + letter grade, 78 checks across
8 categories, plus a **GEO** sub-score — visibility in ChatGPT / Perplexity /
Google AI Overviews), audit_diff (compare vs the previous snapshot),
robots.txt & sitemap.xml validators, JSON-LD & meta/OpenGraph builders, and
per-check fix explainers. Probes TLS, redirects, TTFB, AI-crawler access (GPTBot)
and llms.txt. Free tier + **PRO+** (deep-crawl, GEO, 40+ deeper checks) — API
keys unlock the deeper engine on the **PRO+** plan and up (see Free vs PRO+).
Why pair it with an LLM?
A language model on its own infers a site from training data and, at best, one
rendered page — it can't directly read your TLS certificate's expiry, measure
response time, parse sitemap.xml, or check whether GPTBot is blocked in
robots.txt. This server runs those checks for real: actual HTTP requests,
security headers, redirect chains, structured-data validation — and returns a
**deterministic** score (same site → same number), reproducible enough to put in
a client report. Think of it as the instrument and the LLM as the analyst that
interprets the readout — the two work best together.
What you get
- **audit_site(url, depth=1, lang="ru")** — one call returns:
- health score 0–100 (higher is better) and a letter grade A–F;
- issues grouped by category (security, SEO & indexing, performance, GEO, …),
each with status pass / warn / fail;
- a short human-readable summary.
- **audit_diff(url, lang="ru")** — re-audits a site and compares it to the
previous run: health/score delta and which checks got worse or better. The
first call stores a baseline. This is something a one-off LLM question can't
do — track a site over time. Stateful feature — needs a key on the **PRO+**
plan or higher.
- **list_checks(lang="ru")** — the full catalogue of engine checks grouped by
category, with a free / PRO badge on each — so you (and the agent) can see
exactly what the free tier covers and what PRO unlocks.
- **explain_issue(check_id, lang="ru")** — a deep-dive on a single check:
why it matters and how to fix it. Pass a check_id from audit_site or
list_checks. Title and category are localized; detailed advice is currently
in Russian (the advice_lang field reports this).
- **validate_robots(content, lang="ru")** — paste a robots.txt and get back
syntax issues, whether a Sitemap: directive is present, whether CSS/JS is
blocked from render bots, and which AI crawlers (GPTBot, ClaudeBot, …) are
explicitly blocked. No fetch — validates the text you provide.
- **check_sitemap(url, lang="ru")** — fetches a sitemap by URL and reports its
format (urlset / sitemapindex), URL count and common problems (404,
non-XML content type, http:// links, missing , the 50k-per-file
limit). The fetch is SSRF-guarded on the server.
- **build_jsonld(type, fields, lang="ru")** — generates a ready-to-paste
schema.org JSON-LD (Organization, LocalBusiness, Article, Product,
FAQPage, BreadcrumbList, WebSite) and tells you which required/recommended
fields are missing. It never invents data — only what you pass in.
- **build_meta(fields, lang="ru")** — generates meta tags (title,
description, canonical, Open Graph, Twitter Card) and validates the title