An SEO MCP server connects to Claude Code through one pasted config. Once connected, it reads your site, finds what search engines and AI assistants cannot understand, and proposes the exact text changes to fix each issue. You see the before and after, approve the ones you want, and the fix goes into your project file — ready to deploy in the same session. No plugin to install, no separate dashboard to log into.
Three kinds of SEO MCP server exist: data API servers, Search Console servers, and fix-and-audit servers. Each answers a different question. If you want SEO fixes applied in the coding session you are already in, the third kind does that job.
What an SEO MCP server does in Claude Code
MCP (Model Context Protocol) lets Claude Code call outside tools in a live session. An SEO server on that connection gives Claude Code a way to inspect your site and act on what it finds — all without leaving the editor.
Data API servers return numbers from a third-party index: keyword search volumes, backlink counts, domain scores. The right tool when you need to decide what pages to build or which topics to target. They produce data; they write nothing to your project.
Search Console servers surface your own site's performance: impressions, click-through rates, page positions from your Search Console account. The right tool when you want to see how existing pages are doing. They read; they fix nothing.
Fix-and-audit servers scan your live site for issues that stop search engines and AI assistants from reading it correctly. They give Claude Code the exact text changes needed — page titles, meta descriptions, schema markup, link anchor text — and apply each change to your project files after you approve.
The three kinds answer three different questions. Many teams run a data server and a fix-and-audit server together: one for research, one for execution. The comparison table below sets out what each kind can and cannot do, so you can choose before committing to a setup.
Setup in 3 steps
RankBrain AI is a fix-and-audit SEO MCP server that works with Claude Code, Cursor, Codex, and any tool that accepts an MCP connection.
- Sign up and copy your setup prompt. After creating an account at rankbrain.app, the dashboard shows a one-line setup prompt that includes your API key. No plugin to install.
- Paste it into Claude Code. Open your MCP settings in Claude Code and paste the config. The connection confirms in seconds.
- Ask Claude Code to audit your site. Provide your site's URL and ask for a full SEO audit. RankBrain AI reads the site and returns a prioritized list of fixes.
The same key and one-paste setup also cover Meta Ads, Google Ads, Business details, and Short Reels checks. One connection, one key, every audit type RankBrain AI supports.
Run the first audit
Once connected, type a prompt like:
Audit the SEO of https://example.com and list the fixes.RankBrain AI checks the site and returns a list. Each item shows what is wrong and what the fix is. A typical list includes items like:
- Page title: the current title does not describe what the business sells; a rewrite is proposed
- Meta description: the description is missing or truncated in search results; a replacement is drafted
- Schema markup: no structured data block exists for the business; the full JSON-LD is prepared
- Internal link text: anchor text is generic throughout; specific text is suggested for each link
Nothing is written to your project at this stage. The list is a proposal.
Approve a fix
You review the before and after for each proposed change. When you approve one, RankBrain AI writes it into your project file. Your site looks identical to visitors. Search engines and AI assistants can now read the information they were missing.
To show how the approval step works: a site with no business details in its <head> would receive a proposed LocalBusiness JSON-LD block as the fix — the name, type, address and opening hours drawn from the existing page content. The approval screen shows the current state (empty schema) and the proposed replacement (the full block). Approving it writes the block into the file. Rolling it back is git revert if you use version control.
The fix is in the file after approval. No separate publishing step. No copy-paste from a report into an editor. The pricing page shows before-and-after examples of the fix step — meta titles, schema blocks and link text — labelled as illustrative.
What the plan includes
RankBrain AI has one plan: unlimited SEO audits and fixes, a library of expert plays across SEO, Ads, Business decisions and Short Reels, and support across all tools that accept an MCP connection.
The trial is free for 3 days. No card is required during the free period. After the trial, the plan is $5 a month. Cancel anytime from your account settings.
The plan covers every audit type: SEO, Meta Ads, Google Ads, Business details and Short Reels. All are available through the same key and setup prompt you paste once into your AI coding tool.
Comparison table (checked 2026-11-16)
A data server and a fix-and-audit server address different stages of work. Using both is common.
Other AI coding tools
RankBrain AI works with Claude Code, Cursor, Codex, Gemini, Grok, Windsurf, ChatGPT, OpenCode, Kiro, Cline, and others. The setup prompt is the same for each. The full list is on the pricing page.
If you work in Cursor or Codex, the steps are identical to Claude Code: paste the config, run the audit, approve the fixes.
Safety and common questions
Every fix goes through an approval step. You see the current content and the proposed replacement before anything is written. You approve each fix separately.
The changes are plain text edits to your project files — the same kind of edit you would make by hand. If you use git, every applied fix appears in your repository as a standard diff. Roll back any fix with git revert. No change is applied without your approval.
Will it overwrite my existing content? RankBrain AI shows the full before and after for every change. You approve before the write happens. The file receives exactly what was shown in the approval step.
Does it give a report and stop there? RankBrain AI gives you the exact text change and applies it after approval. The result is a live edit in your project file. The fix lives in the codebase, ready to deploy.
What if I want to undo a fix? Approve only the fixes you agree with. Skip any item in the list. If you approve a fix and later change your mind, git revert removes it in one command.
When a data server is the better pick
A fix-and-audit server works on what is already on your site. It does not answer keyword research questions, assess backlink strength, or tell you which new pages to create.
For keyword research, topic planning, or authority analysis, a data API server is the right tool. The two kinds cover different stages of the same job: use a data server to decide what to build, then use a fix-and-audit server to make sure search engines can read what you have built.
If you want to explore your site's data before connecting an MCP server, rankbrain.app/tools has free SEO tools — a site checker, keyword research, schema generators and more — with no sign-in required. For a longer comparison of approaches, see SEO MCP server vs SEO SaaS and MCP server vs browser extension for SEO.