SE Ranking Remote MCP + GA4: What We Built Live at the Workshop

In February we ran a workshop on Claude Code and SE Ranking's local MCP. At the end of that session we put up a slide with two words: remote MCP. No explanation, just a teaser.

Last week was the follow-up. This post is the written version of what we built live — the GA4 integration I demoed, the skills layer Geoffre walked through, and the parts worth stealing for your own workflow.

What changed with remote MCP#

The original SE Ranking MCP required Docker, a terminal, and a JSON config file. A working install took real effort. If you hit an error mid-setup — and most people did — you needed a developer mindset to debug it.

The remote version removes all of that. You go to your SE Ranking account, find the MCP URL under Profile > API, copy it, and paste it into Claude Desktop as a custom connector. That URL already has your auth embedded. No API keys to manage separately, no local process to keep running.

Two things before you start:

1. Go to Settings > Customize > Connectors > Add custom connector, paste the URL, hit Connect, and authorize via OAuth.

2. Once connected, set read-only tools to “Always allow.” This stops Claude from interrupting long workflows with permission prompts. Leave the 72 write tools at “needs approval” — for the same reason you don’t give an intern unreviewed commit access.

That’s the full setup. SE Ranking exposes 117 read tools and 72 write tools in every Claude Desktop chat from that point on.

Skills: what they actually are#

Having 100+ tools available and knowing how to chain them are different problems. This is where the skills layer comes in.

A skill is a reusable prompt that chains SE Ranking tool calls in a specific sequence and synthesizes the output into a specific deliverable. Think of it as a named workflow: you type /seo-content-brief, Claude knows which endpoints to call, in what order, and what the finished output should look like. You don’t write a new prompt every time.

The full library lives on GitHub and installs in one click: Claude Desktop > Co-work > Customize > Add Marketplace, then paste the repo URL and sync.

After that you have 24 skills across six categories: Content, Per URL, Competitive, Backlinks, AI Search, and Strategy. Anything you do every week as an SEO is probably in there.

One worth understanding in detail: SEO Plan. It’s the top-level skill. You point it at a domain and ask for a quarterly roadmap, and it orchestrates the other skills to produce it. It doesn’t invent the plan — it runs the underlying workflows, reads the outputs, and synthesizes recommendations. The skills aren’t 24 disconnected tools. They’re a system where the higher-level ones call the lower-level ones.

A live run: SEO Content Brief for notion.so#

During the workshop, Geoffre ran the brief skill live for notion.so on the topic “meeting notes templates” in the US market. The tool call sequence:

1. Domain overview, top organic competitors, top organic keywords — in sequence, to establish market position.

2. SERP analysis, related keywords, question-based keywords, keyword gap against competitors — fired in parallel.

3. AI Overview check — which LLMs cite this topic, where Notion sits in the leaderboard.

4. Competitor page fetches: Microsoft Word, Canva, Confluence — the current top three. If a fetch fails, Claude retries automatically.

5. Synthesis: the skill writes the brief and saves output files.

The output includes a primary keyword strategy with volume, difficulty, and intent; secondary keyword clusters; a full H1/H2/H3 outline; and a specific recommendation on whether to create a new page or refresh an existing one. In this case it flagged that Notion already has a URL targeting this topic and recommended refreshing that page rather than creating a new one.

A writer gets the file, ignores Claude entirely, and ships the refresh. That’s the deliverable.

Model note: skills chain multiple tool calls and synthesize across all of them. They reward a bigger model. Sonnet 4.7 or Opus 4 both produce quality output. Haiku will cut corners. Credit consumption is roughly 10,000 to 30,000 SE Ranking credits per content brief run depending on domain size — that’s 3 to 10 runs from 100,000 free credits.

GA4 + SE Ranking: one prompt, one dashboard#

This is the part I demoed. The problem it solves is one every SEO has: SE Ranking has your ranking and visibility data, GA4 has your traffic and conversion data, and combining them is always a manual export-spreadsheet-Data Studio loop. By the time the dashboard is ready, the data is already a day old.

What I showed is a single prompt that pulls both sources and outputs a self-refreshing artifact — a live dashboard inside Claude that updates as you refine it.

The GA4 connection#

Claude has a built-in Google Analytics integration, separate from SE Ranking. You connect it through Claude Desktop’s Integrations panel, authorize via OAuth with standard read-only GA4 access, and Claude can query it directly in natural language. No GA4 API knowledge required, no service accounts, no JSON files.

One request — “show me sessions from organic search for the last 90 days broken down by channel” — translates into the correct GA4 API call. The data comes back inside the same chat where you have SE Ranking data.

What the combined prompt does#

The prompt I ran against a real Ukrainian e-commerce domain does five things in one go:

1. Pulls top competitors and top keywords from SE Ranking.

2. Checks AI search performance across ChatGPT, Perplexity, Gemini, and AI Overviews — brand mentions, share of voice versus competitors.

3. Builds a competitor leaderboard in AI search.

4. Pulls GA4 sessions from AI referral sources for the last 90 days versus the previous 90 days.

5. Identifies the top three pages to update first, with specific reasoning for each.

That last part is the one that actually changes what you do next. Not just “update these pages” — but: page X ranks fourth for keyword Y, gets 12% of your AI referral traffic, Competitor Z is cited three times more often in AI Overviews, here’s what their page has that yours doesn’t.

That kind of synthesis would take hours to compile manually. The prompt produces it in about 10 minutes.

The artifact is live. If something looks off or you want a different date range, you ask Claude to adjust it. It updates in place. You can also export it to PDF or copy the HTML into your own tooling.

What’s on the roadmap#

Two things Geoffre flagged that are worth knowing about.

Multi-user MCP. Right now the MCP URL is tied to an individual SE Ranking account. Agencies with multiple analysts need one URL per analyst. The team-level MCP is in development: one URL, multiple users, per-user usage tracking, account-level billing.

MCP apps. This is the bigger shift. Right now MCP lives as tools and Claude renders text. What’s coming is an interactive application surface with SE Ranking’s branding and workflow logic baked in — like the live artifacts from the GA4 demo, but as a first-class app you can hand to any stakeholder without them needing to reprompt anything.

How to start#

If you weren’t at the workshop and want to set this up:

1. SE Ranking account with API access. New accounts and trial accounts get 100,000 free credits. Older legacy plans may need support to enable it — email [email protected].

2. Claude Pro or Team plan. MCP tools aren’t available on the free plan.

3. Claude Desktop app. The skills install lives in Co-work, not in the browser version.

4. Connect the MCP: Profile > API in SE Ranking, copy the URL, add it as a custom connector in Claude Desktop Co-work > Customize > Connectors.

5. Install the skills: Co-work > Customize > Add Marketplace, paste the GitHub repo URL from the SE Ranking team.

6. For GA4: Claude Desktop Integrations, connect Google Analytics.

Start with one skill. The content brief is the lowest-friction starting point because the output is easy to evaluate. If the brief is clearly better than what you’d write manually, the setup paid for itself.

One thing I’d add from running this day-to-day#

The remote MCP solves the install problem. The skills solve the prompt-engineering-as-second-job problem. What neither of them solves is knowing which workflow to build for your specific site.

For discovery — figuring out what report you actually need and what the right sequence of tool calls looks like — Claude and the skills are the right environment. For running that same report against 50 client domains every Monday, move it to n8n or Claude Schedules. Discover in Claude, automate in your orchestration layer.

That’s the boundary that matters most when you’re deciding where to invest setup time.