The concept: one brain, multiple arms

Claude Code is the brain. The MCPs (Model Context Protocol) are the arms: each one connects Claude to a real-world tool (your browser, Search Console, a keyword tool, your WordPress site). Once connected, Claude no longer just “talks” about SEO—he does it: he reads your actual data, writes content, previews it, and publishes it.

• MCP Chrome DevTools — Claude opens a browser and views the HTML rendering before publication
• MCP Search Console (via Windsor.ai) — Your Site's Actual Impressions, Rankings, and Clicks
• MCP Keywords (Ubersuggest / Ahrefs) — Search Volume, Difficulty, Intent
• MCP imagegen — generates images for articles
• WordPress Post — via the REST API (application password)

That's the core. In total, we connect 12 MCPs —the complete list is right below.

The 12 SEO MCPs to Connect to Claude

You don’t need 10 SEO tools at €200 each per month. You needONE solid SEO infrastructure, connected to your actual data. Here are the 12 MCPs we connect for our clients—and how each one fits in.

The basic idea: Claude Code is in charge. Surrounding it are MCPs (connectors) that give it access to the data. You ask, it fetches the data, and it executes the command. The step-by-step details are right below.

What you do with each MCP—and what you earn

The image shows you the roadmap. Here, MCP by MCP, arethe specific actions you take and the benefits you gain from them.

Install Claude Code

Two ways to get started. The easiest: the Claude app (Mac, Windows, or the web at claude.ai/code)—just open your project and connect the connectors with a few clicks, without touching the terminal. Or use the command line (below), which is more flexible for MCPs running locally.

If you haven't already, install Claude Code (Node.js required), then run it from your project folder.

# Installation (one-time)
npm install -g @anthropic-ai/claude-code
# Run in your project
cd my-project
claude

MCPs are declared either in an .mcp.json file at the root of the project or using the ` claude mcp add` command. Both methods are used below.

MCP Chrome DevTools

The most important thing for quality: Claude opens a real browser, loads the page, and checks how it renders (screenshot + accessibility tree). No more “blind publishing”—we check the HTML before going live.

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest", "--viewport", "1440x900", "--isolated"]
    }
  }
}

Equivalent in a single order:

claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest --isolated

MCP imagegen

To ensure that each article is published with its images, we connect a generator (in this case, OpenAI's gpt-image-1 ). You'll need an OpenAI API key.

{
  "mcpServers": {
    "imagegen": {
      "command": "npx",
      "args": ["-y", "imagegen-mcp", "--models", "gpt-image-1"],
      "env": { "OPENAI_API_KEY": "sk-…ta-clé…" }
    }
  }
}

Search Console via Windsor.ai

To make data-driven decisions (impressions, positions, clicks), connect Google Search Console. The easiest way is to use the Windsor.ai connector, which integrates GSC (and over 300 other sources) into MCP on the claude.ai side.

Claude then retrieves the raw data, which you process into a query to identify the quick wins (pages ranked 5–20 that can move up to page 1):

python3 scripts/gsc_normalize.py \
  --in .cache/gsc-raw.json \
  --out .cache/gsc.json

Ubersuggest or Ahrefs

For search volume, difficulty, and intent, you can connect a keyword tool to MCP. Both Ubersuggest and Ahrefs offer a connector—the same connection process as Windsor.

WordPress Publishing (REST API)

You can publish directly to WordPress via its REST API using an app password (which can be revoked without exposing your real password). Create it in WordPress: Users → Profile → App Passwords.

# Never committed
WP_SITE_URL=https://ton-site.com
WP_USER=your-username
WP_APP_PASSWORD=xxxx xxxx xxxx xxxx xxxx xxxx

A small script posts the content (article or page), which is saved as a draft by default for proofreading:

python3 scripts/wp_publish.py --type post \
  --title "My Title" --slug my-title \
  --content-file article.html --status draft

Skills: Solidify Your Methods

A “Skill” is a written method that Claude follows exactly the same way every time (role + rules + format). This is what makes production consistent and scalable. Each step in the pipeline has its own: keyword research, expert writing, lead magnet…

---
name: expert-writing
description: Writes an in-depth, well-sourced SEO+GEO article.
---
# Rules
1. Research BEFORE writing (data, SERPs, sources)
2. Each paragraph = one piece of data / one example / one method
3. Go into the technical “how,” not the “what”
4. Publish as a draft; never make up numbers

En Claude Code, un dossier ~/.claude/skills/<nom>/SKILL.md est auto-découvert. Tu invoques ensuite le skill par son nom.

Monitoring & Reporting

The loop is completed with monitoring. The MCP Search Console continuously updates a dashboard: what’s rising, what’s stagnating, and where to reinvest. A monthly report is generated automatically, and keyword rankings are tracked via the MCP Keywords tool.

• Positions & Impressions: GSC in Real Time (Windsor)
• Tracking target keywords: Ubersuggest / Ahrefs
• Monthly report: generated and formatted automatically

How It All Fits Together

Each step feeds into the next: GSC data reveals quick wins, keywords inform expert writing, Claude previews the final result before publishing, WordPress receives the linked and illustrated article, and GSC measures the results—which kickstarts the cycle all over again. One brain, one loop.

The 5 Combinations That Change Everything

An MCP on its own is just another tool. The real power comes from the chains: three or four MCPs calling each other within a single prompt, each passing its result on to the next. That’s when the work stops being manual—no more exporting, no more copying and pasting, no more open tabs “just to check.”

Here are the five channels we use most often, along with the exact prompt and what happens behind the scenes.

Take my pages ranked 8 through 15 over the last 28 days.
For the 5 with the most impressions: find out what the top 3
cover that I don't, write the missing sections,
and update the draft page.

Create the brief for “GEO Paris Agency”: keyword expansion,
primary intent, what the top 5 results cover, the H1 structure that outperforms them,
and add it to the Notion database with the status “To be written.”

Write the article for the “X” concept brief using the expert-writing skill,
generate the visuals for the specified locations, publish as a draft,
then open the preview in Chrome and check how it looks.

Submit my 20 target search queries to Perplexity, identify which sites are cited,
compare them to my Google rankings for the same queries,
and log everything in Notion.

Put together the monthly report: Search Console vs. M-1 trends,
GA4 organic conversions, PostHog drop-off points,
and write it up in Notion.

What does it look like in the terminal?

Here’s how Combo^No.1 actually plays out. A sentence is entered; Claude selects his tools, calls them up in order, and produces an output. None of these steps is controlled manually.

> Take my pages ranked 8th through 15th and optimize the top 3.

⏺ Search Console · 28-day search queries
  └ 1,284 queries analyzed · 47 URLs ranked 8th–15th

⏺ DataForSEO · Live SERP (×3)
  └ Top 3 results retrieved for “SEO audit,” “technical audit,” “website audit”

⏺ Firecrawl · Crawl (×9)
  └ 9 competitor pages · Hn sitemaps extracted

⏺ Analysis
  └ /se-audit/: 6 missing sections (including “how much does it cost,” “timelines”)
  └ /technical-audit/: 4 missing sections
  └ /site-audit/: 3 missing sections

⏺ Copywriting · missing sections only
  └ 13 sections · internal links added to 5 pages

⏺ WordPress · wp_publish.py --id 4812 --status draft
  └ ✅ draft updated · slug unchanged

Done — 3 pages enhanced, waiting for you to review them.

How this changes things, in hours

The benefit isn't that "Claude writes faster." The benefit is that data collection, cross-checking, and uploading are eliminated. All that's left are the two things that truly require a human brain: deciding what to focus on, and validating the output.

These are rough estimates based on our own processes for a site with about 100 pages. Your figures will vary depending on the volume and the number of connected MCPs.

Once you're connected, you no longer have to do SEO "by hand"

The infrastructure doesn't "explain" SEO—it runs it on your actual data. Once the MCPs are in place, you can simply ask Claude:

• "Find my Search Console quick wins (positions 5–20) and sort them by traffic potential"
• "Write an article about [keyword], with links to my existing pages, and show me a preview before publishing."
• "Am I mentioned by ChatGPT / Perplexity in my 10 target prompts? If not, why not?"
• "Check this page's speed and Core Web Vitals, and fix any blocking issues"
• "Publish this draft in WordPress, including the generated and compressed images"
• "Generate the monthly report: what's rising, what's stagnating, and where to reinvest"

There are three ways to connect an MCP—you can choose whichever one works best for your stack:

Frequently Asked Questions