zaDojo

Search Console MCP setup errors and fixes

Fixes for the errors community Google Search Console MCP servers throw: API not enabled, 403 permission, sc-domain, access blocked, invalid_grant, spawn uvx ENOENT.

By Denis Veisllari

Most Google Search Console MCP errors come from five places: the Google Cloud project, the Search Console property, the OAuth consent screen, the credentials file, and the AI client's config. Find your error message below, fix that one step, and restart the client. If you would rather skip Google Cloud entirely, the last section shows the route that needs only a Google sign-in.

Google has not published an official Search Console MCP server. Its MCP repository covers Google Analytics, BigQuery, Maps and others, but not Search Console, so every Search Console server you can install is a community project. The two most used are mcp-gsc (installed as mcp-search-console) and mcp-server-gsc. The errors below apply to both, and to most others.

Quick lookup

What you seeLikely causeJump to
"has not been used in project ... before or it is disabled"Search Console API is off in that Cloud projectAPI not enabled
403 "User does not have sufficient permission for site"Service account not added to the property, or wrong URLPermission errors
"is not a verified Search Console site in this account"Domain vs URL-prefix property mix-upWrong property format
"Failed to add user: email not found" in Search ConsoleGoogle bug with new service accountsEmail not found
"Access blocked" or "Error 403: access_denied" on the Google screenOAuth app in Testing, you are not a test userOAuth screen errors
invalid_grant about a week after it workedTesting-mode refresh tokens expire after 7 daysWorks, then stops
"Error 400: redirect_uri_mismatch"Web client used where a Desktop client is neededOAuth screen errors
"Service account credentials file not found"Relative path, or wrong packageCredentials file
spawn uvx ENOENT or spawn npx ENOENTThe client cannot find the commandClient errors
No tools appear and no log is writtenPython older than 3.11Client errors
Numbers do not match the Search Console dashboardData delay, row limits, anonymized queriesNumbers do not match

The API is not enabled

The error is a 403 saying the Google Search Console API "has not been used in project" followed by a project number, often with a link to enable it.

The API has to be enabled in the same Cloud project that owns your OAuth client or service account. The usual mistake is enabling it in a different project.

  1. Compare the project number in the error with the project of your credentials JSON.
  2. In Google Cloud, open APIs & Services > Library in that project and enable Google Search Console API.
  3. Wait a few minutes and try again. Google says new settings can take a moment to reach its systems.

Permission errors

The error is a 403 "User does not have sufficient permission for site". URL inspection says "The caller does not have permission" instead.

Creating a service account does not give it any access to Search Console. Its email has to be added to the exact property you query.

  1. Open the service account's JSON key and copy the client_email value.
  2. In Search Console, open Settings > Users and permissions > Add user, paste the email and give it Full permission.
  3. If you sign in with OAuth instead, use a Google account that already sees the property in Search Console.

If the account has access and you still get a 403, the property string is probably wrong. Read the next section.

The property format is wrong

The error is "'sc-domain:example.com' is not a verified Search Console site in this account", or the same 403 as above.

Search Console has two kinds of property, and the API wants the exact string for the one you have:

  • Domain property: sc-domain:example.com
  • URL-prefix property: the full URL with protocol and trailing slash, such as https://www.example.com/

http and https are different properties, and so are www and the bare domain. Don't guess. Ask the agent to run the server's list-sites tool and copy the site URL exactly as it comes back.

Search Console says the service account email is not found

When you add the service account as a user, Search Console replies "Failed to add user: email not found."

This is a Google problem with newly created service accounts, not something you did. It was first reported around April 2026, came back in September 2026, and also affects Google Analytics. Google has acknowledged it but has not given a fix date. Users report three workarounds:

  • Wait a few hours to a day and add the user again.
  • Add the user through the API instead of the Search Console screen.
  • Use OAuth instead of a service account, if your server supports it.

OAuth screen errors

"Access blocked" with "Error 403: access_denied". Your consent screen is in Testing and the Google account you signed in with is not a test user. In Google Cloud, open Google Auth Platform > Audience > Test users and add your own address.

"Google hasn't verified this app". This shows when an unverified app asks for Search Console access. If you created the Cloud project yourself and only you use it, you can continue past the warning. Verification only matters if you hand the app to other people.

"Error 400: redirect_uri_mismatch". You created a "Web application" OAuth client, but local servers sign in through a loopback address such as http://localhost with a random port. Create a new OAuth client of type Desktop app, download its JSON and point the server at it. For mcp-gsc that is the GSC_OAUTH_CLIENT_SECRETS_FILE setting.

It worked, then stopped a week later

The error is invalid_grant, usually "Token has been expired or revoked."

Google expires refresh tokens after 7 days when an external app's consent screen is still in Testing. Tokens also die if you revoke access or leave them unused for six months.

  1. In Google Cloud, set the consent screen's publishing status to In production. For personal use it can stay unverified.
  2. Delete the token file the server saved.
  3. Sign in again.

If you use mcp-gsc, upgrade too. Versions before 0.4.2 could fail to refresh read-only tokens and delete the token file.

The credentials file is not found

The error is "Service account credentials file not found."

  • Relative path. Tools such as uvx run the server from a cache folder, so a path like ./key.json points nowhere. Use the full path in the client config. On Windows, double the backslashes: C:\\Users\\you\\key.json.
  • Wrong package. For mcp-gsc, install mcp-search-console, not the unrelated mcp-gsc package on PyPI. Run it as uvx mcp-search-console@latest so you get the current version.
  • Wrong file in the wrong setting. A service account key contains "type": "service_account" and a client_email. An OAuth client file starts with an installed or web key. Each goes in its own setting; mcp-gsc uses GSC_CREDENTIALS_PATH for the first and GSC_OAUTH_CLIENT_SECRETS_FILE for the second, and mcp-server-gsc uses GOOGLE_APPLICATION_CREDENTIALS.

The client cannot start the server

spawn uvx ENOENT or spawn npx ENOENT. Desktop apps don't load your terminal's PATH, so they can't find the command. Put the full path in "command". Get it with which uvx on macOS or Linux, or where uvx on Windows. For npx on Windows, use "command": "cmd" with the arguments ["/c", "npx", "-y", ...].

No tools, no log file. mcp-gsc needs Python 3.11 or newer and fails silently on older versions. Run it through uvx, which picks a suitable Python for you, or install Python 3.12.

Config changes do nothing. Check the JSON for trailing commas and unescaped backslashes, then quit the app completely and reopen it. On a Mac, closing the window is not enough; use Cmd+Q.

Where the logs are:

  • Claude Desktop on macOS: ~/Library/Logs/Claude. mcp.log has connection failures, and mcp-server-<name>.log has that server's own errors.
  • Claude Desktop on Windows: %APPDATA%\Claude\logs.
  • Claude Code: run claude mcp get <name> for the error, or /mcp inside a session.
  • Cursor: open the Output panel and pick MCP Logs.

One more limit: local servers run only in desktop and terminal apps. They don't work on claude.ai in the browser or in ChatGPT, which need a remote server.

The numbers do not match the dashboard

These are not errors, but they cause most "the agent got it wrong" reports.

  • Delay. Search Console data usually arrives after two to three days, and the newest days can still change. Some servers ask only for final data, so their last days look empty.
  • Row limits. One API request returns at most 25,000 rows, and Google exposes at most 50,000 rows per day per search type. Long-tail queries on a large site get cut off.
  • Missing queries. Rare and personal queries are left out, so query rows never add up to the total clicks. For exact totals, ask for data without the query or page dimension.
  • Different totals. The chart totals by property and the table by page, so the two can differ. Check the date range too; hourly data uses Pacific time.
  • Quotas. URL inspection allows 2,000 checks a day per site, so batch inspections on a big site stop partway. Search Analytics allows 1,200 queries a minute per site.

Skip Google Cloud entirely

Every error above except the last section comes from running your own Google Cloud project. zaDojo's Google Search Console MCP is hosted for you: you connect Search Console in a zaDojo project with a normal Google sign-in, and your agent reads it at https://app.zadojo.com/mcp. There is no Cloud project, API to enable, service account, consent screen or credentials file, and it works in claude.ai and ChatGPT as well as desktop clients.

Search Console reads, URL inspection and index coverage are free on zaDojo's Free plan, which needs no card. So are Google Analytics reads and site audits up to 50 pages. Keyword, competitor and AI visibility data use credits. To compare it with the other free options, see free SEO MCP servers.

Questions people ask

Is there an official Google Search Console MCP server? No. Google publishes MCP servers for Google Analytics and other products, but not for Search Console. Every Search Console server is a community project or a hosted service.

Should I use a service account or OAuth? A service account suits a server that runs unattended, but you have to add its email to every property. OAuth uses your own Google account and sees every property you already have, but in Testing mode you have to sign in again every 7 days until you set it to In production.

Why does my agent see only some of my sites? The account or service account it uses can only see properties it was added to. Ask the agent to list sites and compare that list with Search Console.

Can a Search Console MCP server change my site? No. These servers read and manage Search Console data. None of them edits your pages.