How we run Walma's marketing on MCP: Ahrefs, Search Console, GA4 and Google Ads in one session
We connected our marketing tools to Claude through MCP and wrote 46 guides in three languages in a few days. Here is the stack, the workflow, and why the real gains only appear when the tools can talk to each other.
A week ago walma.ai ranked for two keywords. Both came from old landing pages that no longer exist. Today the site has a guide section with 46 pages in Swedish, English and German, every page built on real search data, with hreflang, FAQ schema and author blocks. It took a few working days, not a few months.
What made the difference was not the model. It was that the marketing tools were connected.
The problem with marketing in ten tabs
A marketing team lives in a dozen systems that do not talk to each other. Keyword data in Ahrefs. Rankings in Search Console. Behaviour in GA4. Spend in Google Ads. Tags in Tag Manager. Customers in the CRM. Every question that spans two of them, "which keywords drive the sessions that actually convert, and what do we pay for them", means three exports and a spreadsheet.
AI assistants have not solved that. A chat that cannot reach your systems can write a text but cannot know whether it is needed. This is where MCP, the Model Context Protocol, changes things: the standard that lets an assistant use tools. With one MCP server per system, the same session can ask all of them.
The stack we run
This is what our marketing runs on today, all through Claude Code and Walma AI Hub:
- Ahrefs MCP for keyword data, difficulty, traffic potential and competitors' rankings. Ahrefs' own official server.
- Search Console MCP for what we actually rank for and what gets clicked. Our own server, with a service account.
- Google Analytics MCP for what visitors do afterwards. Google's official server.
- Google Ads MCP for spend, search terms and quality scores. Google's official server for reads, our own functions for writes.
- Google Tag Manager MCP for auditing the container and consent settings.
All of them sit behind AI Hub in our own Azure tenant in the EU. Credentials live in one place, every call is logged, and write tools require approval. The same setup our customers get.
How it went
Day one: research. With Ahrefs connected, we asked Claude for keywords tied to our offering across three markets. Not "AI" in general, but the phrases people actually type: "what is a mcp server", "claude code installieren", "was kostet chatgpt", "seo för ai sök". Ranked by traffic potential divided by difficulty, not by volume. The result was a list of 60 ideas, prioritised.
Days two and three: writing. Each guide was written with the search data alongside, headings phrased as the questions, and the other guides as link targets. Because Claude Code sat in the same repository as the site, everything landed in the right folder with the right frontmatter. The build prerendered it all.
Day four: verifying. Search Console through MCP confirmed the old pages were on their way out and nothing cannibalised. Ahrefs' SERP overview showed which domains we compete with on each term. The Tag Manager server confirmed the consent tags were wired correctly before the traffic arrived.
The method is written up in Best MCP servers for marketing and SEO.
The gains come when the tools talk to each other
Each server saves time on its own. The big win is the combination.
An example from last week. The question: "Which of our news articles get impressions in Google but few clicks, and what would a better title be worth in ad spend if we bought the same traffic?" Search Console gave the pages and their CTR. Ahrefs gave the CPC for the keywords. GA4 gave what visitors from those pages did. Google Ads gave what we actually pay for adjacent terms. One answer, one session, twenty seconds.
Before MCP that was a morning's work nobody did, because it was not worth a morning. Now it runs every week, because it costs nothing. That is the real efficiency gain: not that the same work goes faster, but that work that never got done gets done.
The next step is packaging it as skills: "weekly SEO review", "search terms to negate", "AI-search visibility check". Then anyone on the team runs the same routine, with the same tools, and gets the same quality.
What it takes to do the same
Three things, in order:
- The servers. Most already exist. Google, Ahrefs, HubSpot, Salesforce and Shopify have official ones. Our guide to the best MCP servers for marketing lists them, and the setup guides for Claude Code, Cursor and Claude Desktop show how to connect them.
- The governance. An ad account, a CRM and a Tag Manager are the most expensive credentials in the company after finance. They should not live in environment variables on laptops. A gateway that holds them centrally, gives the right people the right tools, requires approval for anything that spends or publishes, and logs every call. That is what AI Hub does.
- The routines. Skills that describe how you do recurring tasks, so the result is the same no matter who asks.
If you want to see what it looks like in practice, with your own accounts, a 20-minute walkthrough is the fastest way. Book one here.