Google Ads MCP: connect Google Ads to Claude or ChatGPT, four ways compared

Google Ads shows you what happened. An AI chat can help you spot patterns in it and suggest changes for you to review, but only if it can see your numbers. This guide compares four ways to get Google Ads and GA4 data into Claude or ChatGPT, from a free CSV export to a connector that can also change your campaigns. The example throughout is a check worth repeating every month: search terms that spent money but recorded no conversions. Prices and features come from each vendor’s own pages as of 2 November 2026; we read the documentation, we did not run the connectors on a live ad account.

What is the easiest way to analyze Google Ads data with AI?

• Once a month, one account: export a search-terms report as CSV and upload it to the chat. Free, no setup, but the data is a snapshot.
• Only Google data, and someone on your team is comfortable with a terminal: Google’s own MCP servers for Google Ads and Google Analytics. Free and open source; they read data and do not change campaigns.
• Several data sources in Claude or ChatGPT, without code: Windsor.ai. Free for one data source and one account; $19 a month billed yearly ($23 monthly) for three. It can also change campaigns if you let it.
• You want a Supermetrics connection only for the AI chat: Starter from $44 a month billed yearly, for one user and one destination.

What is the Google Ads MCP server?

MCP, short for Model Context Protocol, is the standard that lets an AI chat call an outside tool, such as a data connector, while it answers you. A Google Ads MCP server is that tool for your ad account: the chat asks it for campaigns, costs or search terms, and it fetches them from the Google Ads API.

• Google’s own server (googleads/google-ads-mcp on GitHub) is free and open source. It runs on your computer, needs Python, a Google Cloud project and a developer token with Explorer access, and lists no tools that change campaigns. A sibling server does the same for Google Analytics, marked experimental.
• Hosted MCP servers such as Windsor.ai skip the setup: you sign in and add a connector in Claude or ChatGPT. They cover more platforms, and some, Windsor.ai included, can also change campaigns.

What can you ask once the data is connected?

The value is not a prettier report. It is questions that would take an hour of filtering in Google Ads:

• Wasted spend: “Which search terms cost more than $20 in the last 30 days and recorded no conversions? Group them by campaign and suggest negative keywords. Analysis only, do not change anything.”
• Ads versus website: “Show Google Ads clicks and GA4 sessions from google / cpc side by side for each campaign. They are not the same metric; flag differences above 20% and list what I should check.”
• Budget pacing: “Project this month’s spend from spend to date and compare it with my target of $1,500. Show the formula.”
• Across platforms: “Cost per conversion for Google Ads and Meta Ads, last quarter versus the one before. First list the conversion definition and attribution setting each platform uses, and say if they are not comparable.”

Treat every answer as a draft. Check two or three figures against the Google Ads interface before acting on them. A search term without conversions in 30 days may still convert later, and if conversion tracking is broken, the AI sees zero conversions everywhere.

The four ways compared

How to do it for free with a CSV export

1. In Google Ads, open Insights and reports, then Search terms, and pick the last 30 days.
2. Add the columns Cost, Clicks, Conversions and Campaign, and download the table as CSV.
3. Open the file once before uploading. Search terms can contain names, phone numbers or addresses that people typed into Google; delete those rows.
4. Upload the file to Claude or ChatGPT and ask the wasted-spend question above. Ask for the rows each suggestion is based on, so you can check them.

The limit: next month you repeat every step, and combining Google Ads with GA4 or Meta Ads means three exports that the AI has to match up.

How to connect Google Ads MCP to Claude with Windsor.ai

1. Connect Google Ads at Windsor.ai. Sign up at Windsor.ai (affiliate link) and connect your Google Ads account. The 30-day trial allows 10 data sources and 15 accounts. After that, the free plan keeps one data source and one account, with the last 30 days of data. Google Ads plus GA4 are two data sources, so that combination needs Basic after the trial.
2. Add the connector in Claude. Open Customize, then Connectors, search for Windsor and click Add, then sign in with your Windsor.ai account. In ChatGPT, Windsor.ai is listed as an app.
3. Set the permissions before the first question. See the next section.
4. Ask a precise question. Name the period, the threshold and the output you want. The wasted-spend question above is a good first test.
5. Make the change yourself, or approve it in the chat. Negative keywords can be added by hand in Google Ads, or pushed by Windsor.ai after you approve the change.

Can Claude or ChatGPT change your Google Ads campaigns?

With Windsor.ai, yes. According to its server reference, the AI can create, pause and enable campaigns, ad groups and ads, set budgets and bidding, manage keywords, push negative keywords and manage Customer Match lists in Google Ads, and pause campaigns or change budgets on Meta, TikTok, LinkedIn and Microsoft Ads. All of this runs through one tool called execute_action, on every plan, including the free one.

Windsor.ai instructs the AI to confirm with you before it uses that tool. That is an instruction to the model, not a lock. Its Claude setup guide also suggests choosing “Always allow” for quicker access. We recommend the opposite:

• Never “Always allow” the write tool. In the connector’s tool permissions in Claude, let reading tools run and set execute_action to ask for your approval every time.
• Read the approval request. It should name the account, the campaign, the current value and the new one. If it does not, decline and ask.
• For analysis only, connect a read-only login. Give Windsor.ai a Google Ads user with read-only access and test once that a change is refused. Writing “analysis only” in the prompt helps, but it does not block anything.
• Check the change history. After every approved change, look at Change history in Google Ads, and make one change at a time.

You can re-enable a paused campaign, but not the clicks and leads it missed in the meantime. A budget raised by a misunderstood prompt costs money until someone notices.

How much does each option cost?

• CSV export and Google’s servers: no software cost. Your time is the price: a CSV is out of date the next day, and Google’s servers need a Google Cloud project, Python and, for Google Ads, a developer token.
• Windsor.ai: free for one data source and one account. Basic is $19 a month billed yearly ($23 monthly) for three data sources, 75 accounts and unlimited users. A fourth data source means Standard at $99 a month billed yearly ($118 monthly). The terms say all sales are final, with no refunds, including for prepaid monthly fees.
• Supermetrics: Starter is $44 a month billed yearly ($55 monthly) for three data sources with three accounts each, one user and one core destination, plus 4,000 AI credits a month. If you already use that destination for Looker Studio, check whether the AI chat needs a second one. There is a 14-day free trial.

The AI chat is extra: you need a Claude or ChatGPT account, and long data questions use up a plan’s limits faster. All Windsor.ai plans, including agency examples, are in our Windsor.ai review.

Where does your data go?

With a CSV upload, the report goes to the AI provider under your chat account’s settings. With a connector, two companies are involved: the connector fetches the data from Google, and the AI provider receives what the connector returns for your question.

Windsor.ai says it runs its servers in Germany, caches data for at most about three weeks and keeps it out of backups. Its terms include a data processing agreement with EU Standard Contractual Clauses, and its sub-processor list names US companies such as Cloudflare (logs) and SendGrid (email). Google’s MCP servers run on your own computer, so no extra company stores the data in between; your AI provider still receives everything the server returns.

Ad reports are commercially sensitive, and search terms can contain personal details. Customer Match lists and CRM data are personal data — check the points in AI tools and the GDPR before you connect them.

Which one should you pick?

• Try the idea for free first: a CSV upload of last month’s search-terms report shows in ten minutes whether the answers help you.
• Google Ads and GA4 only, and you have a developer: Google’s MCP servers. Free, and the Google Ads server has no write tool.
• Several data sources, no code, or agency accounts: Windsor.ai. Start on the trial, then the free plan or Basic, and never set the write tool to “Always allow”.
• Supermetrics: worth a look if you want its AI credits and Studio, or already use it for reporting and the destination question is solved.

Sources

• Windsor.ai pricing and plan FAQ
• Windsor.ai MCP server reference
• How to integrate data into Claude with Windsor.ai
• Windsor.ai terms of service
• Windsor.ai security and data retention
• Google Ads MCP server on GitHub
• Google Analytics MCP server on GitHub
• Supermetrics pricing