To use SiteLine in Claude.ai, add it as a custom connector: paste the server URL, approve the OAuth screen on SiteLine, and repeat for the second server. There is no token to copy. This page covers the Claude.ai connector flow. Claude Desktop and Claude Code have their own pages, and the category and usage guides are linked at the end.
Which Claude are you using
| Product | Credential | Page |
|---|---|---|
| Claude.ai (web) | OAuth connector | this page |
| Claude Desktop | OAuth connector, or a bearer token in a local config | Claude Desktop Search Console MCP |
| Claude Code (CLI) | Bearer token | Claude Code Search Console MCP |
The data is the same imported Search Console and cookieless analytics in each case. The credentials are separate stores: a connector approved in Claude.ai does not appear in Claude Code, and the reverse. Create both if you use both.
Add the connector
- Copy
https://siteline.tools/mcp/search-consolefrom Settings → MCP access in SiteLine. - In Claude, add a custom connector and paste the URL. Menu labels change; look for Connectors.
- Complete the SiteLine approval screen. The app then appears under Connected apps in SiteLine.
- Repeat for
https://siteline.tools/mcp/analyticsif you want pageviews and goals. - Revoke under Connected apps when the project ends.
SiteLine registers OAuth clients dynamically, so you do not mint a client id or paste a secret. Access is scoped to the team of the approving user, not to the Claude workspace name. For Claude for Work or Team, a custom connector is still subject to that workspace's admin rules. The Google scope is read-only (webmasters.readonly), so the connector cannot submit sitemaps or change Search Console.
If you added only Search Console, Claude has no pageview data. If you added only analytics, it has no queries. Ask for the server you mean by name.
What Claude.ai is good for
Claude.ai has no repository. Use it as a briefing surface: name the query, the URL and data_through, write the narrative, then move to Claude Code or Cursor to make the edit. Or paste the briefing into the pull request yourself.
First prompts to try:
list-sites, then get-import-status, then get-site-overview for dates through data_through. Then get-striking-distance. Web only.
list-analytics-sites, then overview and a page breakdown for the same closed dates. Visitors are not clicks.
Put the standing rules in a Claude Project's instructions as well: totals from get-site-overview, never average position or CTR, never sum rows, and do not equate visitors with clicks. The server instructions already say this, but a user asking for "a table of query clicks that adds up" can talk a model out of it.
FAQ
Can one Claude conversation see both servers?
If you added both connectors, yes. Claude still picks the tools. If it only calls Search Console, it has no pageviews.
Does Claude Code pick up my Claude.ai connectors?
No. Different credential stores. Use a bearer token for the CLI.
Will Claude change rankings?
Not through SiteLine, which is read-only. If you grant Claude other write-capable MCP servers (a CMS, GitHub), that access is separate.
Keep reading
- Google Search Console MCP: what a Search Console MCP is and what SiteLine exposes.
- MCP setup for agents: config for every other client.
- How to query Search Console with MCP: the call order and the pitfalls.