Google Ads MCP — Official vs GoMarble Hosted

Google's Google Ads MCP launched a day before Meta's, but the two could not be more different in shape. Where Meta shipped a hosted MCP with 29 tools and write access via Facebook Business OAuth, Google shipped an open-source repository with three tools and read-only access. The Google Ads MCP is built around running GAQL queries (Google Ads Query Language) against the API — if you know GAQL, you can write any analysis you want. If you don't, the three tools (list accessible customers, run a search query, get resource metadata) are a thin layer. There's no write access, no scheduled jobs, no managed hosting — you clone the repo, configure OAuth credentials, and run the server locally in Claude Desktop's MCP config. For engineering teams who want full control and are comfortable writing GAQL, this is the right primitive. For paid media operators who actually want to pause campaigns, adjust budgets, push changes, or get analysis without writing query language, the official Google Ads MCP is the wrong surface entirely — not because Google built it wrong, but because it's optimized for engineers, not operators.

What is the Google Ads MCP, and what is it actually like to use?

Google released the Google Ads MCP on 2026-04-28 as part of Google Ads API MCP Server. It's a Model Context Protocol server that lets Claude, ChatGPT, Cursor, and other MCP-compatible clients call the Google Ads API directly from a conversation — no scripting, no custom integration code, no separate workflow tool.

Auth: Google OAuth via the Google Ads API. Access mode: Read-only — no write operations exposed. Pricing: Free (open source). The server exposes 3 tools across 3 capability areas: list_accessible_customers (list accounts you can access), search (run GAQL queries against the API), get_resource_metadata (introspect Google Ads resources).

Google's official Google Ads MCP is a developer-first server, and that fact matters for any honest comparison. The architecture is open-source on GitHub at googleads/google-ads-mcp, the three exposed tools (list_accessible_customers, search, get_resource_metadata) are intentionally thin, and the assumption is that the consumer knows GAQL well enough to write any query they need. For engineering teams building internal tools, for developers experimenting with MCP architecture, for technical operators who genuinely prefer query languages over conversational tools, the official server is well-designed for that audience. Auth is via the Google Ads API OAuth flow — standard developer credentials, not the simpler end-user OAuth that Meta's MCP uses. Pricing is free (open source). Where this stops being the right answer is the moment a paid media operator actually wants to operate. The Google Ads MCP is read-only by design — it can pull data via GAQL queries, but it can't pause campaigns, adjust budgets, propose new keywords, or push any changes to Google Ads. Operations have to happen in the Google Ads UI separately. The three tools require GAQL fluency for any non-trivial analysis. There's no managed hosting — you're maintaining a self-hosted MCP server, including OAuth credential refresh, rate-limiting, and uptime. None of those are limitations Google got wrong — they're the natural shape of a developer-first reference implementation. They're just the limitations that disqualify the official server for most paid media operators.

Sources: Google Ads MCP server: Developer integration guide · googleads/google-ads-mcp on GitHub · Announcing official MCP support for Google services

What does GoMarble's hosted MCP add, and when does that matter?

GoMarble's hosted MCP at https://apps.gomarble.ai/mcp-api/sse is a single connection that covers 5 ad platforms (Meta Ads, Google Ads, Bing Ads, LinkedIn Ads, TikTok Ads) plus everything the Google Ads MCP doesn't: competitor research via Meta Ad Library, Shopify revenue reconciliation, Klaviyo email/lifecycle attribution, GA4 + Search Console analytics, and Agent Mode with scheduled jobs.

Auth: Sign up at apps.gomarble.ai, connect ad accounts in Integrations, then add custom connector in claude.ai/settings/integrations. Pricing: Public pricing tiers — see gomarble.ai/pricing. Access mode: Read + write (writes via propose/approve pattern for safety).

GoMarble's hosted MCP takes the operator-first approach. Instead of exposing GAQL as the primary interface, it offers structured tools that map to common paid media workflows: keyword discovery, keyword metrics, campaign management with propose-and-approve patterns, ad group operations, account structure analysis, and full GAQL access where it's the right tool. The server is hosted — no GitHub clone, no OAuth credential management, no local MCP infrastructure to maintain — and it covers Google Ads alongside Meta, TikTok, LinkedIn, and Bing, plus Shopify orders, Klaviyo email, GA4 analytics, and Meta Ad Library competitor research. For engineering teams who specifically want the open-source primitive and are comfortable with GAQL, the official MCP is the right surface. For paid media operators who want to actually operate their Google Ads accounts — pause campaigns, shift budgets, propose new keywords, push changes — GoMarble's hosted MCP has full read+write access via the propose/approve pattern, no GAQL fluency required, and joins Google Ads data with the rest of the paid media stack in one connection. The trade-off is honest: GoMarble is a paid product on public pricing tiers, where Google's official MCP is free open source. For the engineering-team use case, free open source wins. For the paid-media-operator use case, the paid product is the simpler answer because the operations the official MCP doesn't support are the operations that actually run paid media work.

Side-by-side — the rows that actually matter for a decision

The table below makes the trade-off concrete. The first column is what Google's official MCP gives you (read-only, 3 tools, open source, self-hosted); the second is what GoMarble's hosted MCP gives you (read+write, structured tools plus GAQL, hosted, cross-platform). Read down whichever rows match your actual workflows. The honest read across the rows: if you're an engineering team that wants raw GAQL access and is comfortable maintaining a self-hosted MCP server, the official MCP is the right primitive. For paid media operators who need to actually operate Google Ads accounts, every meaningful row tilts toward GoMarble.

Real workflows: where the cross-channel surface earns its keep

The fastest way to see whether the official Google Ads MCP is enough for your work is to look at the workflows that actually run paid media operations. The three below are queries that come up constantly for any team running Google Ads — not edge cases, just the regular work. Each one is single-prompt in GoMarble's hosted MCP because the underlying tools support both read and write operations, and the cross-platform context is there for the cross-channel questions. Each one either requires writing custom GAQL plus going to the Google Ads UI separately to actually act on the findings (using Google's official MCP), or simply isn't possible because the official MCP doesn't support write operations at all.

Look at the last 30 days of search terms from my Google Ads account. Identify queries that drove spend with zero conversions. Propose a negative keyword list and attach it to my top 3 campaigns.

My top-performing Google Ads keyword is 'best workout shoes for flat feet'. Pull the top 5 Meta video ads from competitors in this space and tell me what hooks they're using. Generate 3 Google Ads RSAs and a Meta ad copy variant aligned to the most common hook.

Compare ROAS across my Google Ads campaigns this month vs last month. Propose moving budget from underperformers to the 3 campaigns with the strongest week-over-week growth, with the new daily budget per campaign.

Setup — both options

https://apps.gomarble.ai/mcp-api/sse

Which should you choose?

From paid media teams using GoMarble

The operator's verdict

Here's the operator-tested verdict the way a senior PPC manager or paid media director would frame it: Google's official Google Ads MCP is genuinely useful for the specific audience it's designed for — engineering teams who want raw GAQL access, developers building internal tools, technical operators who prefer query languages over conversational tools, and anyone who specifically wants the open-source primitive for self-hosting. For that audience, the official MCP is the right primitive at the right price (free). For paid media operators who actually run Google Ads accounts day-to-day — making bid adjustments, pausing underperforming campaigns, proposing new keywords, restructuring account hierarchies, running cross-channel analyses against Meta or Shopify revenue — the official MCP is the wrong surface, not because Google built it badly but because read-only with three tools requires GAQL fluency is the wrong shape for the operator job. GoMarble's hosted MCP is the simpler answer for that audience despite being paid, because actually operating Google Ads accounts conversationally without writing GAQL or self-hosting infrastructure is exactly the use case it's built for. The decision here is about whether you're operating accounts or building developer tooling on top of accounts. Two genuinely different jobs. Pick the surface that matches the work.

FAQ

One MCP, every channel.

Connect once to GoMarble's hosted MCP — get Google ads, every other ad platform, competitor research, Shopify revenue reconciliation, and Klaviyo attribution in the same Claude session.