How to connect Google Ads to Claude: 4 ways in 2026
Fastest way: add Caybl.ai in Claude, sign in with Google, and ask about your campaigns, spend and search terms - about two minutes.
Continue with Google · You choose which accounts to connect · Disconnect any time · No card needed
- Try free in Caybl
- Dry-run any change first
- Pro plan for read/write
Below are all four ways that work today, including the free ones, and what each costs you in setup time.
By the Caybl.ai team · Updated 22 October 2026
Quick answer
- In Claude, open Settings → Connectors and add Caybl.ai from the directory (or add a custom connector with
https://api.caybl.ai/mcp). - Sign in with the Google account that already manages your Google Ads account.
- Ask: “Which search terms cost me money last month without converting?”
Prefer to do it yourself? Google’s open-source Google Ads MCP server is free but needs a developer token and a Google Cloud project. Exports and BigQuery work too, with trade-offs covered below.
The 4 ways compared
| Way | Setup time | Live data | Can make changes | Cost |
|---|---|---|---|---|
| 1. Hosted MCP connector (Caybl.ai) | ~2 minutes | Yes | Yes, optional, with dry runs | Pro plan for Google Ads |
| 2. Google’s official MCP server | Hours, plus developer token approval | Yes | No, reporting only | Free (you host it) |
| 3. CSV or Google Sheets export | Minutes, every time | No, a snapshot | No | Free |
| 4. BigQuery transfer + BigQuery MCP | Hours to days | Daily refresh | No | BigQuery storage and query costs |
Rule of thumb: if you want answers today and the option to act on them, use a hosted connector. If you’re an engineer who wants full control and only needs reporting, run Google’s server. If you need one quick analysis, export a CSV. If your team already lives in BigQuery, use that.
1. Hosted Google Ads MCP connector (Caybl.ai)
Claude talks to outside tools through MCP connectors. A hosted connector runs the MCP server for you, so there’s no developer token to apply for, no Cloud project and nothing to install. You authorize with Google’s own consent screen and Claude can query your account from then on, in the Claude web app, desktop app and Claude Code.
Sign in with Google
Use the Google account that already has access to your Ads account.
Add Caybl.ai in Claude
From Claude’s connector directory, or as a custom connector with https://api.caybl.ai/mcp.
Choose read or read-and-write
Changes are a switch on each connected account. Turn it off to stay read-only.
Ask
“Rank my campaigns by cost per conversion, worst first.”
What you get: campaigns, ad groups, ads, search terms, impression share, keyword ideas with volume and bid estimates, and full Google Ads Query Language (GAQL) reports when you need something custom. With changes switched on, Claude can also adjust budgets, pause or enable campaigns, add keywords and negative keywords, and draft responsive search ads.
Trade-offs: Google Ads in Claude needs Caybl.ai’s Pro plan. Google Analytics is free to connect, and the same connection also covers Search Console and Bing Webmaster Tools, so one authorization answers questions across all of them.
Is your Claude Team or Enterprise workspace blocking custom connectors? See what to send your admin.
2. Google’s official Google Ads MCP server
Google publishes an open-source Google Ads MCP server on GitHub. It’s free and it comes from Google, which matters to some teams. The catch is that you run it yourself, and it needs the same setup as any Google Ads API project.
- A Google Ads developer token, applied for from a manager account and reviewed by Google.
- A Google Cloud project with the Google Ads API enabled.
- OAuth client credentials and a refresh token for the account you want to read.
- Python on your machine (or a server), plus the MCP config to point Claude at it.
Expect a few hours if you’ve used the Ads API before, and longer if the developer token needs review. It focuses on reporting, so Claude can read your account but not change it. That suits engineers and agencies with an in-house developer. It suits most marketers less well.
3. Export reports to CSV or Google Sheets
The zero-setup route: download a report from Google Ads (campaigns, search terms, keywords) as a CSV, then attach it to a Claude chat and ask your question. You can also schedule reports to Google Sheets and paste or export from there.
It works well for a single analysis, like a one-off search-terms clean-up. It breaks down when you ask follow-up questions the export doesn’t cover, because every new angle means another download. The data is also frozen at the moment you exported it, and large accounts quickly hit file size limits.
4. Google Ads to BigQuery, then BigQuery to Claude
Google’s BigQuery Data Transfer Service can copy Google Ads data into BigQuery on a schedule. Connect Claude to BigQuery with an MCP server, such as Google’s MCP Toolbox for Databases, and Claude can write SQL against your ad data alongside anything else in your warehouse.
This is the most powerful option for data teams, and the heaviest. You need a Cloud project with billing, the transfer configured and a grasp of the transfer’s table schema. Data refreshes daily rather than live, you pay for storage and queries, and there’s no way to act on what you find.
Google Ads in Claude Code
Claude Code supports remote MCP servers, so the hosted connector works in the terminal too. Add it once:
claude mcp add --transport http caybl https://api.caybl.ai/mcpYour browser opens to authorize, and from then on you can ask about Google Ads from any project, or have Claude Code pull search terms into a script. Add it at user scope to have it everywhere, or project scope to keep it to one repository. More in Google Analytics in Claude Code.
What you can ask
- Diagnose“Which ad groups are burning budget without converting?”
- Find waste“Pull search terms that cost more than 30 with zero conversions this month.”
- Rank“Rank my campaigns by cost per conversion, worst first.”
- Research“Give me keyword ideas for this landing page, with volume and bid estimates.”
- Compare“What changed in performance between this week and last?”
“Rank my campaigns by cost per conversion, worst first.”
“‘Retargeting - Display’ is worst at $22.10 per conversion, nearly 3x the account average of $7.90. The next two campaigns are close enough to leave alone for now.”
Illustrative example - not real account data.
Copy-paste prompts by role
Agencies
- “List every account I can access with its spend and conversions last month, and flag any where cost per conversion rose more than 20%.”
- “Write a five-bullet monthly summary for this client, in plain English, no jargon.”
In-house PPC
- “Which campaigns are losing impression share to budget rather than rank, and what would it cost to recover it?”
- “Find search terms with clicks but no conversions over 30 days and draft a negative keyword list. Do a dry run first.”
Founders and small businesses
- “Am I wasting money on Google Ads? Give me the three biggest leaks.”
- “Is my daily budget limiting results, or would more budget just buy worse clicks?”
Ads and analytics together
- “Which campaigns bring visitors who actually convert on the site, using Google Ads and GA4 together?” (Connect GA4 as well.)
- “Which keywords am I paying for that I already rank for organically?” (Add Search Console.)
Limits, safety and accuracy
There’s no read-only Google Ads scope. Google’s single adwords scope covers reading and writing together, whichever way you connect. So any read-only limit has to be enforced by the connector itself. Caybl.ai enforces the switch on every call rather than trusting the assistant to behave.
Changes can be off
Turn off “Allow changes” and every read tool still works.
Preview a change
Google validates the whole change and reports what would fail, without applying it.
Created paused
New campaigns, ad groups, keywords and ads never start serving on their own.
Every call logged
Your dashboard shows which tools ran and when.
Freshness and conversion lag. Google Ads reports can lag by a few hours, and conversions are credited to the day of the click, so the last few days keep changing. For stable numbers, ask about periods that ended a week or more ago.
Check before you act. Ask Claude to state the date range, account and conversion column behind every number, and to dry-run any change before applying it. Treat its recommendations like a junior analyst’s: usually right, worth a glance.
Connect Google Ads to Claude in two minutes
Google Analytics is free to connect. Google Ads in your own AI needs Caybl.ai’s Pro plan.
Get started free →FAQ
Not on its own. Claude reaches outside accounts through MCP connectors. You can add a hosted Google Ads connector such as Caybl.ai in a couple of minutes, run Google's open-source Google Ads MCP server yourself, or give Claude exported data as CSV files or through BigQuery.
A small service that exposes the Google Ads API to an AI assistant through the Model Context Protocol, the open standard Claude uses for connectors. Claude calls its tools - list campaigns, pull search terms, run a report - and answers from the live data instead of a pasted export.
Yes. Google publishes an open-source Google Ads MCP server on GitHub. It is free, but you run it yourself and it needs a Google Ads developer token, a Google Cloud project with the Ads API enabled and your own OAuth credentials. It focuses on reporting rather than making changes.
With Caybl.ai, changes are a separate switch on each connected Ads account. It is on by default so the connector is useful straight away, and you can turn it off to keep the account read-only. Changes can run as a dry run that Google validates without applying anything, and anything that could start serving - a campaign, ad group, keyword or ad - is created paused unless you ask otherwise.
Usually the date range, the account time zone or conversion timing. Google Ads attributes conversions to the day of the click, so recent days keep filling in for a while, and the API and the interface can apply different default columns. Ask Claude to state the exact date range and conversion column it used.
Google's own MCP server and the CSV route are free but take setup or manual work. Caybl.ai connects Google Analytics to Claude free; using Google Ads in Claude through Caybl.ai needs the Pro plan.
Not with Caybl.ai. You authorize whichever Google account already has access to your Ads account. If it can reach several accounts, Claude gets the full list and asks which one you mean rather than guessing.
Sources
- Google Ads API documentation - developer tokens and access levels
- Google Ads MCP server - Google’s open-source repository on GitHub
- BigQuery Data Transfer Service for Google Ads - Google Cloud documentation
- Model Context Protocol (MCP) - the open standard Claude uses for connectors