Keyword research over MCP
Keyword research your agent does from your repo.
Your coding agent reads what your site sells and proposes the searches your buyers type. QueryScope sizes each one against your own Search Console and remembers what your agent decided.
From $49/mo, everything included · works in Claude Code, Cursor, Cline and Windsurf
Searches proposed for QueryScope's own site, 2026-11-14.
Why start from the repo
A seed you type finds what you already knew.
The usual loop is a browser tab: type a seed, export a list, paste it somewhere, and decide by hand which rows are your business. Your agent already has the one thing that decides that, your code, and it can read it. So the research starts from what the site sells, and the numbers come back to the place where the page gets edited.
Propose, size, read back, remember.
It starts from what your repo says you sell.
A keyword tool starts from a seed you type, so it returns variations of what you already thought of. Here your agent reads the pricing page, the product pages and the comparisons in your repo and sends up to fifty searches a buyer would type, in their words. It also records what the site sells, who buys it and who it competes with, so every later read aims at the right rivals.
up to 50 searches per call, read from the repo
Keyword research for SaaSEach search gets its demand and where you already stand.
Monthly searches, the position your own Search Console reports for it, a rival that ranks for it, and whether it is already in your plan. Long, specific searches are often too small for a volume figure, so they come back as not priced rather than as zero: unknown is not the same as nobody.
searches/mo · your position · a rival · in the plan
Page-level vs query-level dataWith no arguments, it reads back the gaps.
Call it with nothing and it returns the searches you already rank 11 to 30 for, the buyer-intent searches a competitor wins where you never appear, and the searches Google already shows you for that no plan claims. Ask about one search and you get its demand, your position, impressions and clicks, the page aimed at it, and who ranks above you.
close · rivals win · shown but unclaimed
Find your striking distance pagesWhat your agent decides is remembered.
Each search your agent judges is recorded: a target with a score for how close it is to what you sell, not your buyer, or not yet. The plan is written to queryscope.md in your repo, so you read it as a diff. Up to three countries of fifty searches each. The next session starts from that plan, not from a blank list.
target · not my buyer · not yet
Which SEO change to make firstOn a live site
The plan board, on a SaaS we run ourselves.
AppScreenshotStudio's declared searches, read from production: where each one stands and whether it moved.
How the plan is doing
21 May to 25 Sep · arrows are places moved, up is better
On page 1
where clicks happen · was 0 on 21 May
Below page 1
ranking, earning almost nothing · was 2 on 21 May
Not ranking
no reading on 10+ impressions · was 55 on 21 May
Coverage
<1%The whole site earned 528 clicks in 28 days, most of it on searches this plan never claimed.
On page 1
figma app store screenshot template
#9.9
▲3
320/mo
Demand
38
Impressions
2
Clicks
One line in your IDE. Then authorize and pick a site.
claude mcp add --transport http queryscope https://mcp.queryscope.dev/mcp
Works the same in Cursor, Cline and Windsurf. Your IDE opens a browser once to authorize.
Then ask: “what should we go after?”
FAQ
Keyword research over MCP.
What is a keyword research MCP server?
An MCP server your coding agent can call for keyword data, so the research happens in Claude Code, Cursor, Cline or Windsurf instead of a browser tab. QueryScope's keyword tool is one: your agent sends the searches it read out of your repo and gets each one back sized against your own Search Console.
Is it a keyword database like Ahrefs or Semrush?
No. It does not let you look up any keyword for any site, and it has no backlink data. It sizes the searches your agent proposes for your site, reads your position from your own Search Console rather than an estimate, and names rivals that rank for them. If you need a full keyword database, keep the suite.
Why do some searches say not priced?
Volume figures stop below a floor, and the long, specific searches buyers type often fall under it. Those come back as not priced, which means unknown, never zero. Your agent judges them on what the repo says they are worth to the business.
How do I set it up?
Add the server with one line in your IDE, authorize it with Google, and pick a site. Then ask your agent what to go after: keyword with no arguments is the first call on any site.
Will ranking for these searches bring sales?
QueryScope cannot say. It reads your search data and stops at the click. It tells you which searches fit what you sell and where you stand on them; what happens after a visitor arrives is in your analytics.
Run keyword research from your repo.
Connect your site and ask your agent what to go after. It reads the repo first.
7-day free trial · nothing due today · cancel in one click
More of what your agent can ask