Connect an AI assistant with MCP
Model Context Protocol, or MCP, lets an AI assistant read data from another service. Searcherries provides a read-only MCP connection so the AI assistant in Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code or another MCP client can analyze your collected project data and help you improve your website with it.
Use MCP in the Searcherries sidebar to view the server URL and connection instructions. MCP is available with paid-plan access, including an active eligible trial.
What an assistant can read
The connection exposes your projects and, for each of them: the AI visibility metrics of a period (score, mentions by platform, change against the previous period, share of voice), the customer questions and every stored AI answer with its judgments (how prominently the brand is named, sentiment, off-topic answers, evidence quotes), the web searches the platforms ran, competitors with what platforms say about them, cited URLs and domains, the business profile, and the connected data from Google Search Console, Bing Webmaster Tools and GA4 (top queries and pages, opportunities, trends, question-shaped searches, AI traffic and landing pages, monthly history). It can also return a prioritized action plan built from that data.
The data belongs to the Searcherries account you authorize. A connection does not give access to other users' projects.
What the connection can do
The Searcherries MCP server reads data already stored by Searcherries. It cannot edit projects, change questions, delete data, manage billing, or trigger fresh calls to analytics providers.
For example, "Where are we losing to competitors in AI answers, and which pages should we improve first?" can be answered from stored data. "Run a new Google Analytics import now" is not an operation this MCP server offers.
Choose a client
- Claude: use a custom connector and browser authorization.
- Claude Code: add the server from the terminal and sign in with
/mcp. - ChatGPT: turn on developer mode, add the server as an app and sign in through the browser.
- Codex: add the remote server and sign in through the browser.
- Cursor, VS Code and other clients: paste the server configuration and sign in through the browser, or use a personal access token where browser sign-in is not possible.
The remote server address is:
https://app.searcherries.com/mcp/searcherriesThe MCP settings page is at /mcp; the connection endpoint includes /mcp/searcherries. Using the settings-page address as the server URL will not connect the client.
Start with a useful request
After connecting, ask:
What should I work on next for my Searcherries project?
The assistant calls get-action-plan and answers with up to five prioritized tasks. Each task comes with dated evidence, a suggested owner, a success measure and a link to the matching report. Name the brand or domain in the request when the account has several projects.
For a closer look, ask:
List my Searcherries projects, then give me the project summary for the one I select. Where do competitors beat us in AI answers, which pages do the AI platforms cite for those questions, and what should we change on our website first?
The server also provides ready-made prompts that appear in your client's prompt menu once connected: action-plan, geo-audit, content-gaps, competitor-gap, page-brief, question-review and weekly-report. Each one tells the assistant which data to read, in which order, and what to deliver. In Claude Code they are listed as slash commands under the server name.
If the answer requires many records, ask the assistant to retrieve all relevant pages rather than drawing a conclusion from the first page alone. See the tool reference for every tool and its arguments.
MCP and Wizard
Wizard is the built-in, one-question analysis workflow with a monthly Wizard allowance. MCP provides data to an external assistant and does not consume that Wizard allowance. The external assistant may have its own subscription, usage charges, and data-handling settings.
Authorize only a client you intend to use with your project data.