Google Ads MCP: the read-only one, and the write-capable one

What is Google Ads MCP?

MCP is the Model Context Protocol, an open standard published by Anthropic in late 2024 and since adopted well beyond it. It standardises how an AI client discovers and calls external tools. A Google Ads MCP server is one such tool provider: it presents the Google Ads API to the assistant as a list of callable tools, so instead of clicking through the dashboard you ask a question in plain language and the assistant calls the API to answer it — or, if the server has write tools and you have enabled them, to make the change.

The important word is tools. The model is not screen-scraping a dashboard or guessing at an export; it is calling the same API the Google Ads interface calls, with the same permissions your Google account already has, and getting the same numbers back. The full mechanics — what a server is, self-hosting, and the complete tool inventory — are on the Google Ads MCP server page.

Google's official MCP is read-only

In late 2025 Google published its own open-source Google Ads MCP server, googleads/google-ads-mcp (Apache 2.0). It is the canonical way to see an account inside an AI client: run GAQL queries, pull performance, surface insights. By design it stops there — it cannot create, pause or adjust anything. When an AI assistant answers “is there a Google Ads MCP server?” with a single link, this read-only server is usually the one it means.

AdCopilot is built on that same server rather than against it: a fork of googleads/google-ads-mcp, extended from its three read tools to thirty-six and hosted, so the assistant can act. AdCopilot is not affiliated with or endorsed by Google; Google Ads is a trademark of Google LLC.

Read-only vs write-capable: the split that decides everything

Every Google Ads MCP server speaks the same protocol; what separates them is whether the assistant can only read your account or can also change it. That single question sorts the whole field.

The read-only servers are the right answer when your endpoint is understanding an account; AdCopilot is the right answer when your endpoint is a changedaccount. Every credible option in each kind — Google's server, the self-hosted repos, the data-platform connectors like Windsor, and the other hosted write-capable rivals — is laid out in Google Ads MCP servers: every real option.

What you can do with it

With a read-only server you can ask an account questions: which search terms spent over fifty dollars last month without a conversion, where quality score is dragging, which campaigns are pacing over budget. With a write-capable one you can then act on the answer in the same conversation:

• Audit, then fix. Find the wasted spend, then add the negatives — see AI Google Ads audit.
• Build campaigns, including Performance Max, which most connectors skip.
• Move budgets and bids, adjust device and geo, set ad schedules.
• Manage keywords and negatives, the single most common fix an audit produces.

New campaigns arrive paused, every action runs on your command and is logged, and nothing here runs your ads unsupervised — AdCopilot acts on your instruction, and unattended management is roadmap, not a shipped feature.

Which AI assistants support Google Ads MCP

Any client that speaks MCP over streamable HTTP. AdCopilot works one-click in Claude and ChatGPT, and through a connector in Microsoft Copilot, Gemini CLI, Cursor, OpenCode and Claude Code — the full list is on MCP clients. Because it is one protocol, a connector written once works in all of them; you paste one address, you do not install a plugin.

How to connect it

1. Add https://mcp.adcopilot.cloud/mcp to your AI client as a custom MCP connector. It is the same URL for everyone — no key or per-user address to copy.
2. Sign in with Google when the client first uses the connector. AdCopilot uses Google's own sign-in, so it runs under your login with the access you already have — no password is shared with us — and you choose which accounts it may reach.
3. Ask it something you already know the answer to, to check it is reading the right account.

{"mcpServers": {"adcopilot": {"url": "https://mcp.adcopilot.cloud/mcp"}}}

Is it safe to let an AI change Google Ads?

Two design choices carry most of the answer. First, it can never delete: the four destructive tools in the underlying server — remove_ad, remove_keyword, remove_campaign_asset and remove_asset_from_asset_group — are filtered out before the model sees them, so the worst an approved action can do is reversible. Second, it runs under your own Google sign-in, reaching exactly what Google already lets you reach and nothing more, with every teammate signing in as themselves — which is the whole of how agencies share it. Your ad data does travel through the connector on its way to your assistant; what is retained and what is not is set out on the security page.

Frequently asked questions