Search Console data, where the pages are written
Every other way of reading Search Console puts the numbers somewhere other than the thing that produced them. You read a position in a browser tab, form an opinion about a page, and then go looking for the template that rendered it. When you connect google search console to claude code that gap closes, because the assistant answering the question is already sitting in the working directory with the route file, the template and the copy within reach. Reporting and fixing stop being two separate activities divided by a context switch you were never going to make twice in one afternoon.
The connector URL
https://api.mevnix.com/mcp
The same URL for every source. Add it once in Claude Code; it carries whichever accounts you have connected.
What Search Console knows that your repository does not
A codebase is a complete record of what you decided to publish and it says nothing whatsoever about what anybody wanted. Search Console is the other half of that: the queries that produced an impression, how far down the results each one sat, and which URL Google decided to show for it. The second half is the part you cannot reason your way to from the source. A heading you agonised over may be drawing nothing at all, and a sentence you wrote in passing may be the only reason a page ranks, and the sole record of either is a property you have to go and ask.
A terminal that can act on what it just read
A dashboard ends at the number. It shows a page sitting at position nineteen and the next move is entirely yours: open the project, find the file, work out which of several templates produced that particular URL. Running search console in claude code collapses all of that. The session that pulled the row can open the file that rendered it, show you the title tag it currently emits, and draft a replacement you either accept or argue with. Nothing reaches your site without you saying so, because Mevnix only reads and your editor still needs your hands, but the distance between noticing and doing shrinks to a sentence.
The loop only closes when both halves are in one place
The reason a claude code seo workflow earns its setup is the second lap rather than the first. Measure, change something, ship it, wait, measure again: that cycle is where ranking actually moves, and it breaks in practice because the measuring lives in a browser and the changing lives in a repository, so the last step quietly never happens. With both in the terminal you can ask what a page was doing a fortnight ago, what you altered since, and whether the alteration did anything, against a working tree that still remembers the commit that made it. No reporting tool can answer that, because none of them has ever seen your git history.
What that looks like
Which pages take impressions on a phrase their own title never uses?
| Query | Impressions | Position |
|---|---|---|
| headless cms migration guide | 362 | 14 |
| migrate to a headless cms | 118 | 22 |
| headless cms migration checklist | 54 | 41 |
An invented example rather than a customer account. The pattern is what matters: a single page, a single intent, three wordings and three very different positions. The remedy is a title and an opening paragraph using all three naturally, and that lives in a file in your project rather than in a setting somewhere.
Setting it up
- Authorise Google once
Sign in here and connect your Google account. The consent screen asks for read access, and you choose afterwards which properties are exposed, so nothing becomes readable until you have deliberately picked it.
- Register the address on the command line
Add the address above as an HTTP MCP server in Claude Code, under whatever local name suits you. It is a single command that writes to your Claude Code configuration; there is no process to supervise and no credential file to keep out of version control.
- Finish signing in, then ask from inside the project
Use the MCP command in a session to complete the browser sign-in. The tools are then available wherever you start Claude Code, which makes the natural place to ask the repository of the site the numbers are describing.
Free covers one source of each type and three sources in total. $9/month removes the limits on one business; running several is $19 for the first and $9 for each one after it — see pricing.
Questions
- How is this different from the Claude desktop connector?
- The connector and the data behind it are identical. What differs is what the assistant can reach besides them: in the desktop app it has your numbers and nothing else, while on the command line it also has the repository you are standing in, which is why an answer there can name a file rather than describe a symptom. Both are worth having and one account covers both.
- Do I need an API key for this?
- No. Claude Code can complete a browser sign-in, so it uses the same approve-once flow the desktop app does and no long-lived secret lands in a configuration file. Keys exist for the tools that genuinely cannot open a browser, such as scheduled jobs and scripts, and they carry read scopes by default for precisely that reason.
- Can it edit my site or push a change?
- Mevnix cannot. It asks Google for read scopes alone, so submitting a sitemap or requesting indexing sit outside what it is able to do at all. Changing files is your editor doing its own job, under whatever confirmation you already get before something is written or run.
- How fresh are the numbers inside a session?
- Search Console settles its data on a two to three day delay, which is Google behaving as designed rather than anything we do, so a question asked this morning reports a window ending a few days back. Answers are briefly cached so three follow-ups are not three round trips, and anything more than a few hours old is fetched again.