What You Can Do With a Google Ads MCP (Manual Ad Creation Is Over)

I have been running Google Ads for over ten years. I started in the era when the job was mechanical: you changed bids by hand, one keyword at a time, and you built keyword lists in square brackets and quote marks. [exact match] for the terms you knew converted. "phrase match" for the ones you wanted to explore. Broad match I simply did not use, because broad match was a way to donate your budget to Google and find out afterwards.

That discipline worked. It also took hours a week, and most of those hours were spent reading the search terms report and adding negatives, because the whole system only held together if you kept pruning it.

Google has spent the decade since taking those levers away. Close variants ate the precision of exact match. Smart bidding took the bid slider. Performance Max took the account structure. The manual account is mostly gone and it is not coming back.

So here is the part that surprised me. A Google Ads MCP server does not give you the old manual control back, it gives you something closer to what the manual work was actually for: constant, specific attention to the account. The difference is that you are now directing it in sentences instead of clicks.

First, the split that decides everything: read or write

There are two kinds of Google Ads MCP server and they are not interchangeable.

Google publishes its own, free and first-party. It exposes three tools: one to list the accounts you can reach, one to run GAQL queries against them, and one to describe the API structure so the model can write valid queries without you knowing GAQL. In Google's own words, it "is strictly read-only. It cannot modify bids, pause campaigns, or create new assets."

Then there are hosted connectors from third parties, and those write. Ryze AI's exposes 150+ tools with write actions behind approval controls, and it is free to connect. Adspirer creates whole campaigns across six ad platforms and leaves them paused for you to launch. Optmyzr's writes keywords and only keywords, behind a mandatory preview.

Start with the free read-only one. A fortnight of asking questions will tell you whether your actual complaint was ever about making changes, because for a lot of advertisers the honest answer is that they wanted the reporting, not the automation.

What you can do with read access alone

This is the half that costs nothing, and it replaces more of the old job than I expected.

• Mine the search terms report properly. "Pull every search term from the last 30 days with spend over $100 and zero conversions." That query used to be a filtered export and twenty minutes of squinting.
• Find wasted spend across the whole account. "Which campaigns spent the most with the worst CPA last month, and what changed versus the month before?"
• Diagnose a move rather than notice it. "Non-brand CPA rose 30% last week. Break it down by campaign, device, match type and hour of day, and tell me which one accounts for most of the change."
• Audit structure. "List every ad group with fewer than three active ads, or with a single keyword, or with no negative keyword list attached."
• Check match type reality. "For my phrase match keywords, show the actual search terms Google matched them to, sorted by spend." This is the modern version of the discipline I used to run by hand, and it is the single most useful question on this list.
• Build the weekly report. Ask for it in the shape you want, in the order you think, without a dashboard in between.

None of that changes anything in your account. All of it used to be an afternoon.

What you can do once the assistant can write

This is where "manual ad creation is over" stops being a slogan.

• Negative keywords from the search terms report, in one pass. "Find search terms over $100 with no conversions and add them as exact match negatives at campaign level." The loop that took me hours a week for a decade is now one sentence and a review step.
• Build a campaign from a brief. Ad groups, keyword lists with the match types you specify, responsive search ads, budgets. Adspirer creates the whole thing paused so you review a draft rather than discover live spend.
• Prune on a rule you state in words. "Pause every ad group with CTR under 1% and spend over $50 this month." No rule builder, no script.
• Shift budget between campaigns based on a condition you describe, rather than a bid strategy you hope understands your business.
• Fix ad copy at scale. "Find RSAs with fewer than 10 headlines and draft the missing ones in the voice of the existing set."
• Do it across platforms in one conversation. Most connectors cover Meta, TikTok and LinkedIn as well, so "compare CPA by channel and move budget to the best one" stops being a spreadsheet job.

What this does not replace

Manual ad creation is over. Manual thinking is not, and the gap between those two things is where the value moved.

An assistant will happily build a 200-keyword campaign from a thin brief. It has no opinion on whether those are the right 200 keywords, whether your landing page matches the intent behind them, or whether the offer is any good. It cannot tell you that your CPA target is wrong because your sales team closes 8% instead of the 20% in the model. It does not know that half your conversions are bad-fit leads that your CRM quietly discards.

The skill that used to be worth billing for was knowing where to click. The skill that is worth something now is knowing what to ask for, and being able to tell when the answer is wrong. The second one is much harder to fake, which is good news if you have been doing this a while.

The other thing that has not changed: match type discipline still matters, it just moved. You no longer defend precision by refusing broad match, because Google has made that position increasingly hard to hold. You defend it by mining what the system actually matched you to and negating the junk relentlessly. That is the same job I did by hand for ten years. It is now a prompt.

Before you give one write access to live spend

Analytics MCPs are low stakes. The worst case is a wrong number. Ads MCPs spend money, so four things are worth setting before you connect one.

• Check what state new things land in. A campaign created paused is a draft you review. A campaign created live is money already moving. Adspirer leaves them paused. Ask before you let anything build.
• Use scoped credentials where they exist. Pipeboard offers permission-scoped tokens from its Pro plan. A credential that can read performance and adjust budgets but cannot create campaigns is a much smaller mistake than one that can do everything.
• Prefer narrow write scopes to broad ones. Optmyzr writes keywords only, behind a forced preview. That is less exciting and much harder to regret.
• Do not run it unattended. An agent with write access to ad spend should not be on a schedule. Run it while you are watching.

That last one is worth being blunt about. The pitch for autonomous ad management is that it works while you sleep. The reality is that an agent making an unreviewed judgement call at 3am on a live budget is a category of mistake you will find out about from your bank, not your dashboard. Every one of these tools is faster than doing the work by hand even with you sitting in the loop, so the unattended version buys you very little and risks a lot.

How to actually start this week

Two routes, depending on how much you like a terminal.

The free, first-party route. Install Google's own server. It runs locally in Python via pipx and needs a Google Cloud project, a Google Ads API developer token and OAuth credentials. Budget an afternoon. What you get is raw GAQL access to your own data with no vendor interpretation in between, and no ability to break anything.

The hosted route. Add a connector and sign in with Google. Ryze AI is free to connect and needs no developer token: you paste a connector URL into Claude or ChatGPT, authenticate, and you have read and write across 150+ tools with approval controls on the writes. Adspirer installs from Claude's Connector Directory in about two minutes.

Then ask it the question that shows you the value fastest:

textPull every search term from the last 30 days with spend
over $100 and zero conversions. Group them by campaign,
show total wasted spend, and tell me which ones you would
add as negatives and why.

If the number that comes back is uncomfortable, that is the report doing its job, and it is the same report I used to build by hand every Monday for ten years.

For the full comparison of what each tool writes and what it costs, my roundup of Google Ads automation tools covers ten of them grouped by exactly that question. If you have already installed Google's server and hit the read-only wall, what each alternative does that Google's cannot is the same tools sorted by that one problem.

And if MCP itself is new to you, the practical guide to using MCP with marketing tools covers what a server is and what changes once one is wired up, while API vs MCP vs CLI explains when each shape is the right one.

The bracket keys are retired

I do not miss changing bids by hand. I do miss knowing exactly what my account was doing, which is what all that manual work really bought, and which Google's automation quietly took away in exchange for convenience.

A Google Ads MCP is the first thing in years that gives that back. Not the levers, the visibility, and the ability to act on what I see without booking an afternoon for it. Manual ad creation is over. Paying attention is not.