An MCP server for keyword research lets your AI assistant pull real keyword data, such as search volume, related terms, difficulty, search intent and trends, while you ask questions in plain language. Without one, Claude or ChatGPT can brainstorm keywords, but the numbers they attach are guesses. With The SEO MCP connected, the assistant calls live keyword tools (10 in the Keywords category alone, plus keyword ideas, difficulty and intent tools in the Labs category) and answers with data you can check. Setup is one hosted URL and a browser sign-in, and you start with 1,000 free credits.
Last updated: September 2026
The part everyone gets, and the part they miss
Most people understand the pitch: ask your assistant for keywords, get real numbers back. What’s less understood is that keyword research was never one lookup. It’s a chain.
Seed, expand, measure, filter, classify, check the SERP.
In a traditional tool you run each step by hand and export between screens. An assistant without data skips the middle entirely and hands you a confident list with invented volumes. No wonder so many people have stopped trusting AI keyword lists. The real job of an MCP server is to let the assistant run the whole chain with live data at every step. In my view, the quality of a keyword list depends less on any single metric than on whether that chain ran end to end.
What the keyword tools cover, step by step
Here’s what’s on The SEO MCP’s tools page, grouped by where each tool sits in the chain rather than by category.
Expand a seed. Related keywords from Google Ads data, keyword ideas and suggestions from seed terms, semantically related keywords, Google Autocomplete suggestions, and Bing keyword suggestions for a URL.
Start from a website instead of a word. Google Ads keywords for a website, and the keywords a domain already ranks for. Point either one at a competitor and you could start with a seed list already tested by someone else’s traffic.
Measure demand. Google Ads search volume, Bing search volume, estimated ad traffic, Google Trends data, and search-trend breakdowns by subregion and by demographic.
Filter by difficulty. Bulk keyword difficulty scores, plus a keyword overview with metrics.
Classify intent. A search intent classifier, so you can separate people learning from people buying.
Check the SERP. Live Google organic results and a SERP competitors tool, so you see who actually ranks before you commit a page to a term.
Find gaps. Keyword overlap between two domains or two pages. It’s the fastest route to “what do they rank for that we don’t?”
Then there’s AI search volume. One tool reports how often keywords are searched in ChatGPT, Gemini and Perplexity, with a 12-month trend. Frankly, I think it’s the most overlooked keyword metric available right now. Google Ads volume tells you about one audience; this tells you about a different one, and most keyword workflows still ignore it.
Three prompts that run the chain
You don’t call tools by name. Describe the outcome and the assistant picks them.
The opportunity sweep. This one comes straight from our tools page:
“Show me keywords with monthly volume > 500 and keyword difficulty < 30 that semantically relate to ‘invoicing software for contractors’. Include search intent classification.”
That single prompt calls the semantically related keywords tool, then bulk difficulty, then intent classification.
The competitor gap.
“What are the top 5 organic competitors for [your domain]? For each, show me keywords they rank top-3 for that we don’t rank for at all. Group the top 20 gaps by search volume into content topics.”
The regional check.
“Compare search interest for ‘emergency plumber’ across US subregions over the past year, then pull Google search volume for the most common related terms.”
Swap in your own terms. The structure matters more than the wording: say what you want to end up with, and give the filters as numbers.
What it costs to run
Keyword costs vary a lot by tool, so batching matters. On the Growth Engine plan, the tools page lists Google Autocomplete at 3 credits per call, Google Ads search volume at 130 per call (up to 1,000 keywords at once), bulk keyword difficulty at 35 per 100 keywords and keyword ideas at 35 per 100 results. The pricing page adds that actual cost depends on data volume and plan tier: higher tiers use more credits per call, but each credit costs less, so every call costs less in dollars.
The plans themselves:
| Plan | Price | Credits |
|---|---|---|
| Free | $0 | 1,000 once you verify your email, one time |
| Local Dominator | $99/mo | 50,000 per month |
| Growth Engine | $199/mo | 200,000 per month |
| Full Stack | $399/mo | 1,000,000 per month |
All plans include all 93 tools. Credits reset monthly and don’t roll over.
One practical tip. A single prompt can fire 3 to 10 tool calls, so ask for “the top 20” rather than “everything” until you’ve seen what a sweep costs you.
Where this falls short
Two honest caveats.
Search volume from Google Ads data is an estimate of demand. It doesn’t tell you how many of those searches click through to anything, so treat it as a ranking signal between keywords, not a traffic forecast.
And the assistant does the filtering and summarizing, which means it can get that part wrong. Most MCP clients show which tools were called and let you open the raw output. When a number drives a decision, open it. I’d do that for any keyword you’re about to build a page around.
(If you genuinely enjoy scrolling a 5,000-row keyword table with a dozen filters, a dashboard is still nicer for that. No argument here.)
Getting connected
Setup is the same on every client: add the hosted server URL, then sign in through your browser the first time a tool runs. No local install, no API key. Our claude mcp page walks through Claude Desktop and Claude Code, and the chatgpt seo mcp guide covers ChatGPT. Still comparing options? Start with best mcp server for seo or the broader seo mcp server overview.
Frequently asked questions
What is an MCP server for keyword research?
It’s a server that connects an AI assistant such as Claude, ChatGPT or Cursor to live keyword data through the Model Context Protocol. You ask in plain language, and the assistant calls tools for search volume, related keywords, difficulty, intent and trends, then answers with real numbers instead of guesses. The SEO MCP is one hosted example you connect with a URL and OAuth sign-in.
Can ChatGPT or Claude do keyword research without an MCP server?
They can brainstorm keyword ideas and group them into topics, but on their own they don’t have direct access to search volume or difficulty databases, so any figures they give you are estimates at best. Connecting an MCP server gives them tools that fetch current data when you ask, and most clients show exactly which tool produced each number.
Which keyword metrics does The SEO MCP return?
Per its tools page: Google Ads and Bing search volume, related keywords and keyword ideas, keywords for a website or ranked by a domain, bulk keyword difficulty, search intent classification, Google Trends and search-trend breakdowns by subregion and demographic, estimated ad traffic, and AI search volume for ChatGPT, Gemini and Perplexity with a 12-month trend.
How many credits does keyword research use?
It depends on the tool, the data volume and your plan. On the Growth Engine plan, the tools page lists Autocomplete at 3 credits per call, search volume at 130 per call for up to 1,000 keywords, bulk difficulty at 35 per 100 keywords and keyword ideas at 35 per 100 results. Every account starts with 1,000 free credits, and credits on paid plans reset monthly without rolling over.