Know which blog posts to refresh, expand, merge, or kill - without guessing.
A MCP server that turns your scattered SEO and analytics data into one clear verdict per URL. Plug it into Claude, Cursor, or any MCP-aware client and ask: "Which three posts should I update this week?" - and get an answer backed by hard numbers.
What it does
Per-URL SEO verdicts from GSC, GA4, Matomo, Clarity, AI citations: refresh/expand/merge/kill.
with reason codes, evidence, and a 0-1 confidence score. Reporting only - the server never mutates your posts.
Why it matters
Most content teams have analytics in five tabs and a gut feeling. That's how good posts rot quietly, mediocre posts get over-promoted, and the obvious "rewrite this one" is invisible until traffic has already cratered.
This MCP closes the loop:
One question, one URL in, one verdict out.
Same logic across the whole cohort, so the ranking is comparable.
All decisions traceable to numeric thresholds you can pin in src/verdict/rules.ts.
AI clients (Claude, Cursor, MCP hosts) can drive the entire content audit in plain English.
Who it's for
Content marketers running a blog of 50+ posts and tired of guessing what to refresh.
SEO consultants doing audits who want a portable, deterministic scoring layer instead of bespoke spreadsheets.
AI-first content teams wiring up rewrite agents - this MCP is the upstream signal layer.
Indie publishers on Ghost, WordPress, Hugo, Astro, Next, Webflow, or any CMS that exposes a sitemap.
What you get
After one cohort run you have:
A ranked table of every post with a verdict and confidence score.
A markdown brief per "refresh" URL: numbers + top queries + suggested actions an editor (or a writing agent) can act on immediately.
A list of "quick wins": queries sitting at positions 5-15 with below-expected CTR - the fastest title-rewrite wins on the property.
A historical AI-citation diff: which LLMs cited you and stopped.
Every env var is optional. Adapters that lack their env config skip their slice of the snapshot; the server still boots. The verdict engine works on whatever slices are present.
Platform integration
Point it at any site, no CMS plugin required. The post-discovery layer resolves in priority order:
POSTS_LIST - JSON array of {url, title?, published_at?, tags?, word_count?}. Use this when you already have a content index and want exact control.
Ghost Admin API - if both GHOST_ADMIN_API_URL and GHOST_ADMIN_API_KEY are set, Ghost is used as a richer metadata source. Optional.
HTML extraction - per-URL og:title, article:published_time, and JSON-LD datePublished are read live from the URL.
XML sitemap - set POSTS_SITEMAP_URL to your sitemap (or sitemap index) and the server enumerates posts from <loc> + <lastmod>.
Most users only need POSTS_SITEMAP_URL. WordPress, Hugo, Astro, Next.js, Webflow, Framer, Wix, Squarespace, Notion-as-a-site, Substack-mirror sites all expose a sitemap by default.
To add a brand-new platform: nothing to build - just point POSTS_SITEMAP_URL at it.
Tools exposed
Tool
What it returns
posts_list
Posts with {url, title, age_days, tags} from sitemap, Ghost, or your POSTS_LIST.
posts_snapshot
Per-URL unified rollup for a 30/60/90-day window: GSC + Matomo + GA4 + Clarity + citations + meta.
posts_decay_curve
Weekly GSC clicks/impressions/position buckets + a decay/plateau/growth trend label.
Same env vars as the MCP server. --format markdown is supported for cohort_report and posts_refresh_brief; other tools fall back to fenced JSON.
Companion skills + Cursor rule
Three thin routing files ship in the repo so the LLM in your client knows when to reach for these tools:
skills/seo-performance/SKILL.md - tool-routing skill. Drop into ~/.claude/skills/seo-performance/ (or .claude/skills/ per project) to auto-load in Claude Code. Routes a single question to the right tool.
skills/weekly-audit/SKILL.md - one-shot weekly audit playbook. Composes gsc_quick_wins + cohort_report + posts_cite_loss into a deduped, cross-signal ranked digest with proposed edits per URL. Drop in alongside the routing skill.
cursor/rules/seo-performance.mdc - copy to .cursor/rules/seo-performance.mdc in any Cursor workspace.
All optional. The MCP server works without them; they just shorten the "which tool do I call" round-trip.
MCP prompts
The server exposes three prompts that bundle the playbook. Any MCP client (Claude Desktop, Claude Code, Cursor, Continue) can list and invoke them:
Prompt
What it runs
audit_cohort
cohort_report on posts >=90d, then posts_refresh_brief per refresh/expand/merge row. The weekly audit.