Search Console, Google Analytics and keyword data for your AI agent.
dev.seoskill/seo — Model Context Protocol (MCP) Server
The dev.seoskill/seo MCP server provides access to Free Google Search Console (GSC) data, combining “real GSC” with Google Analytics and keyword data so it can be used by AI agents.
🛠️ Key Features
Google Search Console (GSC) access
Google Analytics data access
Keyword data access
Described as “your real GSC” and “Free”
🚀 Use Cases
Supplying AI agents with search performance data
Enabling analysis workflows that combine GSC, analytics, and keywords
⚡ Developer Benefits
Single MCP server interface for GSC, Google Analytics, and keyword data
Supports AI-agent-driven use of search and analytics inputs
⚠️ Limitations
Available information does not specify authentication, supported operations, or tool details.
The SEO data I give my agents every day, so they work from my real numbers
instead of guessing.
Most AI SEO tools do one of two things. They hand your agent a checklist of
generic advice, or they feed it estimates from a keyword tool that has never
seen your site. Either way you end up with the same pages everyone else is
publishing. Pages written from a keyword list aren't SEO. They're slop.
This works the other way round. It's a Google Search Console MCP server and
CLI, and it doesn't tell your agent what to do. It gives it the facts: your
clicks, impressions and positions from Search Console, your visits from
Google Analytics, Clicky or Simple Analytics, your Bing data, the dates of Google updates, and
a crawl of your site. Your agent works out what matters, and you decide what
changes.
It's free and open source. Hack around with it. Make it your own.
Installation (30-second setup)
1. Connect it to your agent
Claude Code
bash
claude mcp add --transport http seoskill https://seoskill.dev/mcp
Add https://seoskill.dev/mcp as a remote MCP server. Cursor, Claude
Desktop, VS Code and most other agents support it.
The CLI (for crawls and Bing, which run on your machine)
bash
npm i -g seo
seo start
seo start signs you in, connects your agents and installs the skill.
2. Sign in with Google
Your agent opens a Google sign-in. Pick the account that can see your Search
Console. That's your account. There's no password and no signup form.
3. Ask it something real
text
Why did clicks drop on example.com last month?
That's it. You're ready to go.
What you can ask it
"Why did clicks drop last month? Was it the core update?"
"Which pages are close to page one, and what do they rank for?"
"Is this page indexed? If not, why not?"
"Which searches does this page show up for that it never mentions?"
"Can ChatGPT, Claude and Perplexity read my site?"
"Who am I up against in search, and is there demand for a 'vs' or alternatives page?"
Your agent picks the commands, reads the results and shows you what it found.
Why it exists
I built this because agents kept failing at SEO in the same few ways.
#1: The agent makes things up
The problem. Ask an agent why your traffic dropped and you get a tidy
list of possible causes. Algorithm update, technical issues, content
quality. None of it comes from your site, so none of it helps.
The fix is to give it your numbers:
seo report shows your totals against the last period, the pages and queries that moved, and the places worth a look: quick wins, low click-through rates, pages stuck on page two, decay and cannibalization.
seo compare lines up the weeks before and after a date, with the Google updates in between.
#2: Keyword tools are guessing too
The problem. Ahrefs and Semrush estimate your traffic from a sample.
They miss most of the long tail, and they rarely agree with each other.
The fix is to start from your own data. Search Console is the source of
truth for Google. When you do need keyword volumes or competitor data,
seo research uses your own
DataForSEO or Ahrefs key, and it keeps those estimates apart from your real
numbers.
#3: The agent says it's fixed when it isn't
The problem. Agents trust their own edits. Then you find the page is
still noindexed, still redirecting, or still not in Google.
The fix is to check the live site:
seo page loads one URL and asks Google whether it's indexed.
seo crawl checks the whole site, including what AI agents can read.
#4: Too many reports
The problem. SEO tools love reports. The first version of this had about
70 of them, and almost nobody used most of them.
The fix was to cut it down to a handful of commands. Each one takes a
site or a URL, and dates if you want them. There are no settings to tune. Every result
ends with the command to run next, so your agent always knows where to go.
Internal links to add across a site or to one page (CLI only)
Every command is also an MCP tool, such as seo_report. Add --json to any
command when you're scripting it.
What it keeps
As little as possible. The hosted server keeps your linked Google accounts
and any keys you save, encrypted. It doesn't keep your Search Console data,
your reports or request logs. The CLI keeps your sign-in in your system
keychain. The privacy policy has the details.