Most SEO questions are the same question with different dates: what changed, why, and what should I do next. The numbers to answer it sit in Google Search Console, Bing Webmaster Tools, GA4 and whatever you use to track AI answers, and the usual routine is to export each one and paste it into a chat. With an MCP connection the assistant reads the numbers itself, with the dates attached, and you only ask the question.
Searcherries has such a connection. It shares everything a project has collected: Search Console and Bing performance, GA4 AI traffic, AI visibility, competitors and cited pages. This article covers what the assistant can read, how to set it up in Claude, Codex and Claude Code, which prompts return something useful, and the routine I now use on searcherries.com itself: an assistant with the data on one side and the website’s files on the other.
What MCP is
MCP, short for Model Context Protocol, is a standard way to let an AI assistant read data from another app. Once Searcherries is connected, Claude can look up your Search Console numbers itself when a question needs them, instead of waiting for you to attach a file.
The Searcherries connection can do two things: list your projects and read their data. It cannot change anything, which is deliberate. You can hand it to an assistant without worrying about what the assistant might do to your project.
Why exports stop working
An export is one report for one period. The questions worth asking cut across reports and periods. “Did the pages that lost Google clicks last month gain AI mentions?” needs Search Console for two periods and the AI visibility report for the same two. “Is the drop on this page a ranking problem or a traffic-source problem?” needs Search Console and GA4 side by side.
You can put all of that together by hand. Most people do it once, for the annual review, and skip it the other eleven months. With the data one question away, the assistant puts it together while you type the next question, and every number arrives with the period it was measured in.
What the assistant can read
Once connected, the assistant can read these parts of a Searcherries project:
- Google Search Console: clicks, impressions, top queries, top pages and top countries for the connected site.
- Bing Webmaster Tools: the same numbers for Bing.
- GA4 AI traffic: visits, engaged sessions, conversion actions and entry pages for each AI source.
- AI visibility: aI Visibility score, brand mentions and share of voice.
- Customer questions and answers: the questions being tracked and the AI answers collected for them.
- Competitors: brands found in the answers, with mentions, average rank and what each platform says about them.
- Citations: the URLs and domains AI platforms cite as sources, per platform.
Long lists come in pages of 25 rows. If you want the whole citations list, say so, and the assistant will keep reading until it has all of it. Every answer carries the date the data was collected, so ask the assistant to mention it: “clicks fell 12% between the June and July snapshots” is a better answer than “clicks fell 12%”.
The connection is available on Starter and Pro and during an eligible trial. It reads what Searcherries has already collected and never starts a fresh import from Google, Bing or GA4. The MCP guide in the docs has the details.
Connect it to Claude
Claude adds Searcherries as a custom connector. It takes about two minutes.
- In Searcherries, open MCP in the sidebar and select the Claude tab. Copy the server URL and the OAuth Client ID shown there.
- In Claude, open Settings, then Connectors, and choose Add custom connector.
- Name it Searcherries, paste
https://app.searcherries.com/mcp/searcherriesas the URL, and paste the Client ID if Claude asks for it. - Continue and approve the Searcherries sign-in screen in your browser.
The Client ID only identifies the connector; the sign-in in your browser is what grants access. Copy it from your own MCP page rather than from a screenshot.
To check the connection, ask Claude to list the projects in your Searcherries account and confirm they are the ones you expected. If a project is missing, the browser probably signed in to a different Searcherries account: disconnect the connector and reconnect with the right one. The Claude setup guide covers the edge cases.
Connect it to Codex
Codex adds the same connection with two commands in the terminal:
codex mcp add searcherries --url https://app.searcherries.com/mcp/searcherries codex mcp login searcherries
The second command opens the browser so you can sign in. After that, “Use Searcherries to list my projects” in a Codex session confirms it works. If sign-in does not go through, the MCP page in Searcherries has an Other clients tab where you create an access token instead. The Codex setup guide shows where the token goes.
Connect it to Claude Code
Claude Code is Claude in your terminal, with access to the files in the folder you open it in. Add the connection with one command:
claude mcp add --transport http searcherries https://app.searcherries.com/mcp/searcherries
Then start a session and type /mcp to sign in through the browser. claude mcp get searcherries should show the connection as connected. If you would rather use an access token, add it when you create the connection:
claude mcp add --transport http searcherries https://app.searcherries.com/mcp/searcherries \ --header "Authorization: Bearer $SEARCHERRIES_MCP_TOKEN"
Prompts that return something useful
The assistant only knows what you ask for, so a good prompt names the data, the period and the shape of the answer you want. Prompts that have worked for me:
- List my Searcherries projects, then analyze the project for searcherries.com. Compare AI visibility, competitors, citations, AI traffic and organic search. Mention the reporting period for each dataset and give me three actions the data supports.
- From Google Search Console, list the queries with more than 100 impressions and zero clicks in the last three months. Group them by the page that earned the impressions.
- From the citations list, show the domains cited for our tracked questions that are not ours, sorted by how many platforms cite them. Which of them publish comparison articles?
- Compare the GA4 AI traffic entry pages with the pages Search Console shows as top pages. Which pages get AI visits but no organic clicks, and vice versa?
- For each tracked customer question, tell me which brand each platform names first and whether we are named at all. Separate what the data shows from what you recommend.
The same style produces whole reports. One request for “an SEO dashboard for the content team with all available data, weak spots and the biggest opportunities” came back as a page with Google and Bing clicks, impressions, a daily clicks chart, traffic sources and top countries, all from one project.
Let Claude Code fix the site
Reports are the smaller half of this. In Claude Code, the assistant that reads your search data can also read and edit the files your website is built from. That turns “which pages should we fix” into “fix them”, with every change shown to you before it goes live.
The setup is the connection above plus the folder with your website’s files. Open Claude Code in that folder, and the assistant has Search Console, Bing, GA4 and AI visibility on one side and your pages, texts and page titles on the other.
This is how the pages this article links to were planned. I asked Claude Code to pull the Search Console data for searcherries.com through the connection and find searches with impressions but no landing page. It found hundreds of impressions for “ai overviews tracker”, “chatgpt visibility tracking” and “brand mentions in perplexity” landing on pages that no longer existed after a redesign, and a cluster of question-shaped searches about tracking mentions in Perplexity with no page at all. The same session then built the tracker pages and the comparison pages, with the copy written against the competitor data the citations report had collected.
Tasks that work well in this setup:
- Find queries in Search Console with many impressions and few clicks, open the pages that earned them, and rewrite the title and meta description.
- Take the question-shaped queries from Content Ideas and draft an FAQ section for the matching page, using the pages from the citations report as the fact base.
- Compare which competitors each AI platform names for a tracked question, then draft a comparison page in the site’s existing article layout.
- Check every page in the sitemap against the Search Console top pages and list the ones that get impressions but are not linked from anywhere on the site.
Two things keep this safe. The connection is read-only, so nothing the assistant does can change your Searcherries project. And Claude Code shows every file change before it saves it, so nothing the assistant writes reaches the site without you reading it. Treat its SEO recommendations the way you would a junior colleague’s: usually right about what to look at, worth checking on what to say.
What it will not do
The connection reads stored data. It cannot run a new AI check, import fresh Search Console data, edit customer questions or touch billing. When the assistant says the latest data is from a week ago, that is the answer; ask it to say so rather than gloss over it. Wizard, the assistant built into Searcherries, has its own monthly allowance; MCP does not use it. The external assistant has its own subscription and data-handling terms, so connect only the ones you use.
Frequently asked questions
Is it safe to connect an AI assistant to my Search Console data?
The Searcherries connection only shares what your account has already collected, you approve it in your browser or with a token you can revoke, and the assistant can only read. It cannot change your project or your Google property. What the assistant does with the data is governed by that assistant’s terms, so check them before connecting a work account.
Which assistants work?
Claude (as a custom connector), Codex, Claude Code and any other assistant that supports remote MCP servers with browser sign-in or an access token.
Does MCP use my Wizard questions?
No. Wizard has a separate monthly allowance. MCP calls do not count against it.
Can the assistant refresh my data?
No. It reads the data Searcherries has already stored. Refreshes happen on Searcherries’ schedule and when you reconnect a source.
Do I need a paid plan?
Yes. MCP is available on Starter and Pro, including an active eligible trial.
Conclusion
An SEO MCP connection is a small thing: one URL and a sign-in. What it changes is the number of steps between a question and an answer, from export, paste and explain down to ask. Connect Searcherries to Claude for analysis, to Claude Code when you want the analysis to end in fixed pages, and ask for the dates on everything the assistant tells you.
