mcpsearchconsole vs mcp-gsc

Both take you from nothing to "show me my top queries", but the work sits in different places.

mcp-gsc, following its Getting Started section as of 2 October 2026:

1. Google API credentials. In Google Cloud Console, create or pick a project, enable the Search Console API, create an OAuth client ID, configure the consent screen as a Desktop app and download the JSON (5 sub-steps). The service-account route is 6 sub-steps, including adding the service account's email as a user on each Search Console property.
2. Install and configure. Install uv (three terminal commands), find the full path to uvx, then paste a JSON or TOML block into your client's config file with the path to your credentials file. Quit and reopen the app.
3. Test. On first use a browser window opens for Google sign-in; the token is then cached on your machine.

The project does not state a setup time.

Search Console MCP, the hosted Google Search Console MCP server:

1. Add https://mcpsearchconsole.com/mcp as a connector in your AI client (in claude.ai it is "Add custom connector"; in ChatGPT it needs Developer mode).
2. Sign in with Google and approve the read-only Search Console permission.
3. Ask a question.

No Google Cloud project, no config file and nothing running on your computer. The trade-off is that your queries pass through our server on their way to Google, while mcp-gsc talks to Google straight from your machine.

What mcp-gsc can do that Search Console MCP cannot

• Run entirely on your machine. Tokens stay in a local file and requests go from your computer to Google. If your policy rules out a third-party server in the path, this is the deciding point.
• Manage properties. It has add_site and delete_site (off by default, enabled with GSC_ALLOW_DESTRUCTIVE=true). We have no property-management tools.
• Batch URL inspection in one tool. batch_url_inspection and check_indexing_issues take a list of up to 10 URLs. With us the AI calls inspect_url once per URL, which comes to the same answers but more calls.
• Service accounts. Useful for unattended scripts. We sign in with a Google account only.
• Be read and changed. The code is MIT-licensed, so you can audit it, fork it and add the tool you need.

What Search Console MCP can do that mcp-gsc cannot

• Work in web chat and cloud agents. claude.ai, ChatGPT and Grok Bot cannot launch a program on your laptop, so they need a remote server. mcp-gsc's own README points web users to its hosted version.
• Ranked analysis in one call. search_insights returns striking-distance queries, low-CTR pages, cannibalization and the biggest movers already scored. mcp-gsc returns analytics rows and period comparisons that your AI then has to analyse.
• Site checks beyond Search Console. DNS, redirect chains, robots.txt, TLS certificates and PageSpeed (Core Web Vitals and Lighthouse).
• Notes that persist. Save a migration date or brand terms once; the next conversation reads them. On Team, notes are shared across the company.
• Read-only by default. We ask Google for the read-only scope; sitemap submission needs a separate opt-in. mcp-gsc requests the full webmasters scope in gsc_server.py (its destructive tools stay off until you enable them).
• Nothing to update. Fixes ship on our side; a local server needs a restart, and occasionally a fix, after dependency changes (its July 2026 release pinned the mcp SDK after version 2.0 broke fresh installs).

What neither does

Neither stores history beyond Google's 16 months, tracks SERP rankings, or reports backlinks or keyword volumes. Both read the same Search Console API, so the numbers match.

About Advanced GSC, the hosted version of mcp-gsc

mcp-gsc's author also sells Advanced GSC MCP, a hosted server with one-click sign-in. On 2 October 2026 its page listed Starter at $15 a month (3 properties), Pro at $40 (10 Search Console and 10 GA4 properties) and Max at $105 (unlimited properties, fair use), plus a limited free version, and named Claude, ChatGPT, Cursor, Windsurf, Zed and Codex. It adds GA4, SERP and keyword research, and on Max backlinks and crawling. If you want those extra data sources in one hosted connector, it is a reasonable choice and costs more than our plans; if you want Search Console only, you are comparing it with Solo at €4.99 a month.

mcp-gsc costs nothing to download. Its real cost is your time: the first setup, plus occasional fixes when a client, Python package or Google token changes. Search Console MCP has a price but almost no time cost.

A simple way to compare over a year:

self-hosted yearly cost = (setup hours + upkeep hours per month × 12) × your hourly rate
hosted yearly cost      = monthly price × 12

Example only, with assumptions you should replace with your own: say setup takes 1 hour, upkeep averages 15 minutes a month, and your time is worth R an hour. That is (1 + 0.25 × 12) × R = 4 × R a year. Solo is €4.99 × 12 = €59.88 a year before VAT, so in this example the two break even at R ≈ €15 an hour. If setup takes you ten minutes and nothing ever breaks, mcp-gsc is cheaper; if you lose an afternoon to OAuth consent screens, it is not.

For a team, each person running mcp-gsc repeats the credential and config work, unless you share one Google Cloud project and its client file. The full plan details are on our pricing page.

From mcp-gsc to Search Console MCP

1. Add https://mcpsearchconsole.com/mcp as a connector in your client and sign in with the same Google account that has your properties.
2. Ask "list my Search Console properties" to confirm access.
3. Remove the gscServer block from your client's config file, so the AI does not pick between two sets of similar tools.
4. Optional: delete the cached token file and, if you made it only for this, the Google Cloud project and its OAuth client.

Plain-language prompts carry over. Saved prompts that name a tool need updating (get_search_analytics becomes search_analytics).

From Search Console MCP to mcp-gsc

1. Follow mcp-gsc's README to create credentials and add it to your desktop or command-line client.
2. Copy anything you saved with add_note (ask "list my notes"), since notes live on our server.
3. Remove the connector from your AI client, cancel any paid plan from your account page, and revoke our access under your Google account's third-party connections.

Note that you cannot use mcp-gsc in claude.ai or ChatGPT on the web, so a move in this direction also means moving those conversations to a desktop client. For every option side by side, see best hosted Google Search Console MCP servers.