Your SEO analyst is now a chat window, not a dashboard

What it does

SEOMonster is a stdio MCP server exposing 70 SEO tools to your AI assistant — Google Search Console, GA4, PageSpeed/CrUX, Cloudflare, IndexNow, and AI-citation tracking. You ask questions in plain English; the assistant calls the tools and reads back the answers. All credentials are yours, resolved at runtime, and nothing goes to a third party.

The interesting bit

The stats work is more careful than the usual SEO tooling. rank_change_attribution uses difference-in-differences against a matched control group with a 95% confidence interval instead of a naked before/after comparison, and it even auto-detects the 2025 GSC data-regime breaks (the impression bug, the num=100 change) and downgrades confidence accordingly. It’s also honest about limits — the README flags that it’s observational, not causal, and that GSC anonymizes ~47% of queries.

Key highlights

• Read-first by default: reads always work; Cloudflare write tools are gated behind SEO_MCP_ALLOW_DESTRUCTIVE plus a per-call confirm token
• content_opportunities ranks topics from your own GSC demand, optionally weighted by GA4 organic conversions, with a transparent scoring formula
• gsc_coverage_audit works around the missing Index Coverage API by bulk-inspecting a URL list and rolling up verdicts
• Lean dependency footprint: stdlib plus the mcp SDK and Google clients; PageSpeed and Cloudflare ride on plain urllib
• Every tool returns the same JSON result envelope, so the assistant gets a consistent shape to reason over

Caveats

• No cold-start keyword research — it prioritizes demand you already have, and the host (not the server) writes any content
• The AI-citation tracking tools are marked optional and the README section is truncated, so their exact data sources are unclear

Verdict

Worth a look if you already have Search Console or GA4 data and live in Claude/Cursor anyway. If you wanted a standalone dashboard or cold-start keyword discovery, this deliberately isn’t that.