Google Search Console MCP server for coding agents

Your search data, where you already work

Search Console keeps the most useful SEO data you own, behind a dashboard. The work happens in an editor and a terminal. Crossing over to click through a chart costs more than the answer is worth, so most people stop checking.

Doverity runs an MCP server that closes that gap. Connect your Search Console property once. After that your agent can ask real questions and get real rows back: which queries bring clicks, where position moved, and why a page is still missing from the index.

Two tools do this. Both read Google's own numbers. Both are read-only. Neither spends credits.

Setup

1. Sign up for Doverity. You get free credits to try the paid research tools.
2. During onboarding, click Connect Search Console and pick the property you want. Doverity holds the Google app, the OAuth client, and the token. You never open the Google Cloud console.
3. Point your client at the Doverity endpoint below. A sign-in screen shows up the first time you connect; approve it once and you are done.

The hosted endpoint is:

https://app.doverity.com/mcp

Claude Code

claude mcp add --transport http --scope user openseo https://app.doverity.com/mcp

--scope user keeps Doverity available in every project. Use --scope local if you want it in this repository only.

Claude Desktop

1. Open Settings, then Connectors.
2. Click Add custom connector.
3. Paste https://app.doverity.com/mcp and approve the login.
4. Custom connectors are not on every Claude plan, so confirm yours allows them before you start.

Cursor

1. Open Cursor Settings, then Tools & Integrations, then MCP Tools.
2. Click New MCP Server. Cursor opens mcp.json. Add:

{
  "mcpServers": {
    "openseo": {
      "url": "https://app.doverity.com/mcp"
    }
  }
}

3. Approve the Doverity login when Cursor asks.

Codex CLI

codex mcp add openseo --url https://app.doverity.com/mcp

Every other MCP client takes the same route: a remote server URL plus a login. Give it the endpoint above and you are connected. Headless and CI setups can use an API key instead. The MCP docs cover that and the usual troubleshooting.

If you would rather wire Google yourself

Every other Search Console MCP server hands you the Google side: create a Cloud project, enable the Search Console API, configure a consent screen, build an OAuth client, download credentials.json, then keep a local server alive. If you want that level of control, self-hosting gives it to you without taking the hosted tools away.

Questions worth asking

Once the endpoint is connected, try these:

• "Which queries put me just outside page one, and what would it take to move them up?"
• "Show pages where impressions climbed but clicks stayed flat. Put the titles and the queries beside each other."
• "What lost the most clicks in the last 28 days? Tell me whether position moved or the demand did."
• "Is /pricing indexed? If not, give me the coverage state and the canonical Google picked."
• "Compare query performance on mobile and desktop for my five biggest pages."

The comparative ones are where an agent earns its place. It can pull two date windows and diff them without a spreadsheet.

What the agent can read

Search performance. Clicks, impressions, CTR, and average position for the property you connected, grouped by query, page, country, device, or date. Sixteen months of history in one call.

URL inspection. Up to 10 URLs at a time. Coverage state, last crawl time, the canonical Google selected next to the one you declared, plus mobile and rich-result verdicts.

That is the whole surface. The agent reads your search data. It cannot publish, delete, or change a setting in your Google account, because the connection requests one scope: webmasters.readonly.

Privacy

Doverity asks for webmasters.readonly and nothing that writes. Query, page, and performance data are used to answer what you asked, and Google's Limited Use requirements govern how that data moves. The Google user data section of the privacy policy lists what is collected, how long it is kept, and how to disconnect a property.

What it costs

Reading your own Search Console data never uses credits. The plan it comes with is $10 a month (₹899 a month in India), which includes $10 of metered usage for the paid research tools — keyword research, rank tracking, backlinks — on the same endpoint. Free credits at sign-up cover the first look. If you would rather not pay at all, self-host the core free forever.

Doverity against the alternatives

FAQ

Is the Search Console MCP read-only?

Yes. Doverity requests the webmasters.readonly scope and nothing that writes. Your agent can pull performance rows and inspect URLs. It cannot submit a sitemap, remove a URL, or change a property setting.

What does it cost?

Nothing on top of the plan. Search Console access comes as part of the $10/month subscription and never draws from your credits. Keyword research, rank tracking, and backlinks run through the same endpoint and spend the $10 of metered usage that comes with the plan. Self-hosting is free.

Which agents work with it?

Anything that speaks MCP: Claude Code, Claude Desktop, Codex CLI, Codex Desktop, Cursor, and any other client that accepts a remote HTTP server. Hand it https://app.doverity.com/mcp and complete the sign-in once.

How fresh is the data?

Google withholds the newest few days of performance until it finishes processing them, so yesterday's numbers can still move. Expect sixteen months of lookback, matching what Search Console shows in its own interface.