An SEO MCP server is a connector that lets an AI assistant such as Claude, ChatGPT or Cursor call real SEO tools in the middle of a conversation: live search results, keyword volumes, backlink profiles, on-page audits. Instead of answering from memory, the assistant fetches the data, then explains it. You add the server once, ask in plain English, and the assistant decides which tool to call. The SEO MCP is a hosted server, so setup is one URL plus a browser sign-in (OAuth). No local install, no API keys, and 93 tools across 12 categories show up in your client.
Last updated: September 2026
The little-known difference: where the server runs
Two SEO MCP servers can list similar tools and still feel completely different to live with. The reason usually isn’t the tool list. It’s where the server runs.
The Model Context Protocol supports two setups. A local server is a program on your own computer that your AI client starts up; you keep it running, keep it updated, and feed it whatever credentials it needs. A remote server is an HTTPS endpoint somewhere else that your client simply points at.
The SEO MCP is the second kind. The server lives at https://mcp.yourdfyhub.com/seo, and sign-in uses OAuth 2.1 with PKCE, with tokens scoped to your own account. Practically, that means:
- Nothing to install or maintain on your machine
- No secret key sitting in a config file on your laptop
- The same account works from Claude, Cursor, ChatGPT and other MCP clients
My take: unless you’re building your own agent and want to own every moving part, a hosted server with OAuth is the right default. The local route is great for tinkering. It’s a chore for someone who just wants keyword data before lunch.
What happens when you ask a question
Say you type “find keywords for coffee shops in Austin.” Here’s the sequence, roughly:
- Your AI client already knows which tools the server offers, because it read the tool list when you connected.
- It picks the tools that fit your request. You don’t name them.
- The server runs those tool calls and returns structured data.
- The assistant reads the results and writes you an answer, often with a table.
Our tools page says a single prompt frequently triggers 3 to 10 tool calls, which the assistant chains together. A competitor-research request, for example, might run labs_competitors_domain, then labs_domain_intersection once per competitor, then labs_keyword_overview to size the gaps.
You can usually watch this happen. Most MCP clients show each tool call inline and let you click in for the raw input and output, which is the fastest way to learn what the server can do (and to catch the assistant taking a detour you didn’t ask for).
What’s actually in the toolbox
The 93 tools are grouped into 12 categories. This is the split listed on the tools catalog:
| Category | Tools | A question it answers |
|---|---|---|
| SERP | 9 | What does page one look like for this query right now? |
| Keywords | 10 | How many people search this, and is it trending? |
| Labs | 18 | Which keywords does a competitor rank for that I don’t? |
| Backlinks | 16 | Who links to this domain, and are any links spammy? |
| On-Page | 4 | What’s wrong with this page, per a Lighthouse run? |
| Domain | 5 | What tech stack is this site built on? |
| Content | 6 | What’s the sentiment around this brand or phrase? |
| Business Data | 4 | What are reviewers saying about this business on Google? |
| App Data | 6 | How does this app look in the App Store or Google Play? |
| Geo-Grid | 1 | How does my Maps ranking change across town? |
| Merchant | 6 | Who sells this product on Google Shopping or Amazon? |
| AI Visibility | 8 | What do ChatGPT, Claude, Gemini and Perplexity say about us? |
Every call fetches fresh data at the moment you ask. Nothing is cached on our side, which is why each one costs credits.
Credits decide your bill, so learn the expensive tools
Every tool call deducts credits, and the cost depends on the tool and your plan tier. On the Growth Engine plan, a Google Autocomplete lookup is 3 credits, a live Google organic SERP is 8 for the first 10 results, a backlink summary is 52, a backlink overlap between domains is 104 per 100 results, and a 5x5 geo-grid scan is 300. Deeper tools cost more because they return more data.
The plans, as listed on our pricing page:
- Free: 1,000 credits once you verify your email, one time, no credit card
- Local Dominator: $99/month for 50,000 credits
- Growth Engine: $199/month for 200,000 credits
- Full Stack: $399/month for 1,000,000 credits
All plans include all 93 tools. One caveat worth knowing before you pick a tier: credits reset every month and don’t roll over, so a quiet month’s leftover balance is simply gone. If you run short mid-month, upgrading your plan adds credits right away; otherwise they refill on your billing date.
Where an SEO MCP server isn’t the right tool
I’ll be straight about this. If your job is hours of manual filtering in a big data table, or handing a polished visual report to a client every month, a traditional SEO dashboard still does that better. An MCP server gives you answers in a chat. The assistant can summarize and build tables, but it isn’t a reporting suite.
Where the MCP approach wins is everything conversational: “what changed last week,” “compare these three domains,” “check this page and tell me what to fix first.” Those are one prompt instead of an export, a spreadsheet and a copy-paste into a chatbot.
Connecting one
The install guides cover Claude Desktop, Claude Code, Cursor, ChatGPT, Windsurf, VS Code with Copilot, and any other MCP client. The pattern is the same everywhere: add the server URL, sign in once in your browser, start asking.
If you’re on Claude, start with our claude mcp page, then read the step-by-step guide to claude mcp seo. ChatGPT users should read chatgpt seo mcp first, since that client has its own quirks. And if you’re still weighing options, our piece on the best mcp server for seo covers what to compare.
Frequently asked questions
What is an SEO MCP server?
An SEO MCP server connects an AI assistant to SEO data through the Model Context Protocol. Once it’s connected, the assistant can call tools for live search results, keyword metrics, backlinks, on-page audits and more while you chat. You describe what you want in plain language and the assistant picks the tools. The SEO MCP offers 93 of these tools across 12 categories.
Do I need to install anything or paste an API key?
No. The SEO MCP is a hosted server, so there’s nothing to install on your computer. You add one server URL to your AI client, and the first time a tool runs you sign in through your browser with OAuth. No API keys are involved at any point. The install guides show the exact line or setting for each supported client.
How much does The SEO MCP cost?
You start free with 1,000 one-time credits and no credit card. Paid plans are Local Dominator at $99/month for 50,000 credits, Growth Engine at $199/month for 200,000 credits, and Full Stack at $399/month for 1,000,000 credits. Every plan includes all 93 tools. Credits reset monthly and don’t roll over, and you can upgrade your plan if you run out.
Which AI assistants work with it?
Any MCP-compatible client. Our pricing page lists Claude.ai, Claude Code, Claude Desktop, Cursor, VS Code with Copilot, Windsurf, and ChatGPT with an MCP connector. You use the same server URL on every platform, and the same 93 tools appear in each one, so you can switch clients without learning a new toolset.
Is the SEO data live or cached?
Live. Every tool call fetches fresh data at the moment you ask, and nothing is cached on The SEO MCP’s side. That’s also why each call costs credits: you’re paying for a current pull, not a stored copy. For time-sensitive questions like today’s SERP for a query, that freshness matters more than it does for slow-moving metrics.