The Model Context Protocol (MCP) is the open standard that lets AI assistants call external tools. Cinble runs a hosted Google Ads MCP server, so you can point an assistant at your account without creating a Cloud project, applying for a developer token or keeping a server alive.
The endpoint
https://google-ads.mcp.cinble.com
Transport is HTTP, and authorization happens through Google's own consent screen the first time you connect: you grant Cinble read access to the customer accounts you choose, and you can revoke it at any time from your Google account permissions or from Cinble.
Early access. We are opening the server in batches alongside the Cinble copilot. Join the waitlist to get your account enabled.
How to connect
Claude (web and desktop)
- Open Settings → Connectors.
- Choose Add custom connector.
- Paste
https://google-ads.mcp.cinble.comand confirm. - Sign in with Google when prompted and pick the customer accounts to expose.
Claude Code
claude mcp add --transport http google-ads https://google-ads.mcp.cinble.com
Cursor, VS Code and other JSON-configured clients
{
"mcpServers": {
"google-ads": {
"url": "https://google-ads.mcp.cinble.com"
}
}
}
Add the block to your client's MCP configuration file and restart it.
ChatGPT and custom agents
Add the same URL wherever your client registers MCP servers. Anything that speaks MCP over HTTP works, including agents you build yourself with an MCP-compatible SDK.
What you can ask once it is connected
You ask in plain language. The assistant picks the tools it needs, queries your account live and answers with the numbers it used:
Three terms took $2,140 with no conversions in the last 30 days:
All three come from broad match on CRM — Generic, which is now 34% of that campaign's spend and 4% of its conversions. Adding them as negatives would return roughly $2,100/month to terms already converting at $54.
Other questions people bring to it:
"Which campaigns increased cost per conversion more than 20% versus last month?"
"Is Performance Max cannibalizing my brand search campaign?"
"Where am I losing impression share to budget rather than rank?"
During early access the assistant recommends the move and leaves the execution to you.
Tools it exposes
| Tool | What it returns |
|---|---|
list_accounts | The Google Ads customer accounts you authorized, with name, ID, currency and manager relationship. |
list_campaigns | Campaigns with type, status, bidding strategy and budget. |
list_ad_groups | Ad groups under a campaign, with status and bids. |
get_metrics | Spend, impressions, clicks, CTR, CPC, conversions, cost per conversion, ROAS and impression share for any entity and date range. |
get_search_terms | The search terms report, so wasted spend is findable without an export. |
get_keyword_performance | Keyword-level results with match type and quality signals. |
compare_periods | Two date ranges side by side, with deltas already computed. |
run_gaql | A read-only Google Ads Query Language query, for questions no prepared report covers. |
Tool names and arguments are discoverable from the server itself, so the assistant learns the surface on connection. The list above is what early access ships with.
Permissions and security
- Read-only during early access. Nothing in the current tool set can pause a campaign, change a budget or add a keyword,
run_gaqlincluded: it accepts queries, never mutations. - You choose the accounts. Authorization is per customer account through Google's own consent screen. Accounts you do not select are invisible to the server.
- Revocable at any time, from your Google account permissions or from Cinble, with no code change on your side.
- Write tools are planned, not default. When they arrive they will be an explicit per-account opt-in with a confirmation step, matching how the Cinble copilot works: it never touches your account on its own.
Hosted versus running your own
Open-source Google Ads MCP servers exist, and they are a fine choice if you want to own the process. The friction is heavier here than on Meta, because Google gates the API itself:
| Self-hosted | Cinble hosted | |
|---|---|---|
| Developer token | Apply through a manager account and wait for Google's approval; a test token cannot see real accounts | Already approved, ours |
| OAuth setup | Cloud project, OAuth client, refresh token | Sign in with Google |
| Manager accounts | You handle login customer IDs by hand | Resolved for you |
| Hosting and rate limits | Yours to keep alive and within quota | Managed |
| Time to first answer | Days, waiting on token approval | Paste a URL |
Where the copilot picks up
An MCP connection answers whatever you think to ask, which makes it excellent for exploration and on-demand analysis. What it cannot do is watch the account while you are not looking.
That is the other half of Cinble: the copilot consolidates every Google Ads and Meta Ads account into one panel, reads their history and puts budget moves in front of you ranked by impact. Both read the same connected accounts, so you can explore in your assistant and let the panel keep watch between sessions.
Get the server enabled on your account
Early access is opening in batches, free while it lasts. Join the waitlist and we will turn on the Google Ads MCP server and the copilot together.
Join the waitlist ↗Frequently asked questions
What is the Google Ads MCP server?
A hosted Model Context Protocol server run by Cinble that connects your Google Ads account to AI assistants, so they can read campaigns, ad groups, keywords and search terms and answer questions about performance in plain language.
Do I need a Google Ads developer token?
No. Applying for one through a manager account and waiting for Google's approval is the slowest part of building your own server. Cinble already holds the approved token, so you only sign in with Google and pick the accounts to expose.
Can the assistant change my campaigns?
Not during early access. The server ships read-only, so no tool can pause a campaign, change a budget or add a keyword. Write tools are planned later as an explicit per-account opt-in.
Can I write my own GAQL queries?
Yes. Alongside the structured tools, the server exposes a read-only GAQL tool, so the assistant can compose a query when your question does not fit a prepared report.
Can one assistant read both Google Ads and Meta Ads?
Yes. Connect this server and the Meta Ads MCP server, and a single assistant can compare both platforms in the same conversation.
How much does it cost?
It is free during early access. Pricing after that will be announced to waitlist members before anything changes.