Your own agent with specialist knowledge

Google Ads MCP: let an AI agent work in your account

An AI agent can go through your search terms every day, exclude waste and prepare new campaigns. Here you can read how that works with a Google Ads MCP, how to connect one yourself and when it is better to have someone with more than 12 years of Google Ads experience build it.

Plug it in. The approval stays with me.
Fabiano points at the Google Ads overview, Pitch steers a mecha with the Google logo and Stack plugs the mecha's plug into the Google Ads screen

By Fabiano · updated November 19, 2026 · 14 min read

Short answer

A Google Ads AI agent works through tasks in your account by itself: fetching data, analyzing it and making or proposing changes. Through an MCP, the connection between an AI and Google Ads, it gets access. Google's official MCP reads your account and creates assets and labels, but adds no exclusions and sets up no campaigns. For that you need a custom MCP, with a specialist's rules built in.

What is a Google Ads AI agent?

An AI agent is software with a language model as its brain. It gets a goal, decides for itself which steps are needed, uses tools to carry them out and checks the result. The difference from a fixed workflow is who chooses the steps: in a workflow they are set in advance in code, in an agent the model chooses them itself (Anthropic).

Agent and MCP: what is the difference?

An MCP (Model Context Protocol) is the plug. It is an open standard that lets an AI safely reach a system, such as Google Ads, Analytics or Meta. The agent is the one that does the work and uses that plug. Without an MCP an agent cannot look into your account. Without an agent an MCP does nothing.

Regular automation in Google Ads

  • Smart bidding strategies, rules and scripts
  • Steps are fixed in advance
  • Suits tasks that are the same every time
  • Cheap and predictable

An AI agent via an MCP

  • Chooses the steps itself for a goal
  • Reads your data via the MCP
  • Suits investigative work that is different every time
  • Has limits and needs a human who decides

What Google already offers itself

Google builds agents into your account itself and at the same time opens the door to external agents. This is the current state, with the source.

Oct 2025Google releases an official MCP server that lets AI assistants reach your Google Ads dataGoogle
Nov 2025Google announces Ads Advisor: an agent in Google Ads that analyzes campaigns and proposes changesGoogle
May 2026Ask Advisor connects Google Ads, Analytics and Merchant Center and makes proposals for campaigns and problemsGoogle
Sep 2026Developer tokens are going away: access to the Google Ads API now runs through your own Google Cloud projectGoogle

Whether Ads Advisor or Ask Advisor is already in your own account, you see in the account itself. Those agents follow the platform's goals. That is not always your margin.

What an agent may do: official or custom MCP

Google's official MCP is a good, safe basis: it reads your account and creates assets and labels after your confirmation (Google). But it only does what you ask, and it does not know what is good for your account. A custom MCP also carries out the work, with a specialist's rules. This is what that difference looks like in practice.

Waste in your search terms

Official MCPShows you the search terms when you ask. Excluding is up to you.
Custom MCP with my rulesLooks every day, excludes clear waste by itself and puts doubtful cases to you.

A new campaign

Official MCPCannot create campaigns, ad groups or keywords.
Custom MCP with my rulesSets up the campaign in a proven structure, as a paused draft.

Your Shopping feed

Official MCPCannot reach Merchant Center.
Custom MCP with my rulesChecks disapprovals, prices, stock and titles daily and prepares improvements.

Ad copy

Official MCPCan create separate texts as an asset, with no check on content.
Custom MCP with my rulesWrites complete ads within your brand rules and checks every promise against your real price.

Who decides what is good

Official MCPThe AI, based on general knowledge and Google's recommendations.
Custom MCP with my rulesMy way of working from more than 12 years of Google Ads, captured in rules.

Changing budget and bids is technically possible with a custom MCP. With me that never happens without your approval. The four MCPs are under MCPs I build.

Connect a Google Ads MCP in 5 steps

Want to start yourself? With Google's official Google Ads MCP server you can do it without programming, if you are handy with a terminal. These are the steps, with Google's guide as the source.

Stack looks at a plug on the blueprint with a question mark, Tally ticks off the checklist and Fabiano points the way
1

Create a Google Cloud project

Turn on the Google Ads API in it. You then get test access. A developer token is no longer needed since September 9, 2026 (Google).

2

Request access to your real account

In the Cloud Console you request Explorer access. It is usually approved automatically and gives up to 2,880 operations per day. Keyword ideas from the Keyword Planner need the higher Basic access, with a brand verification.

3

Install the official MCP server

On your laptop with Python 3.10 or higher, pipx and gcloud. Or as a service on Google Cloud Run, with login through OAuth, so your team can share it.

4

Choose your AI

If you want to ask questions yourself, connect the MCP to Claude, ChatGPT, Gemini CLI or Cursor. If the agent has to run by itself at fixed times, you need a model API, such as the Gemini API.

5

Start small and read only

Turn off writing in the settings file (tools_config.yaml) and give the agent one task, such as a weekly report. Expand only when you trust the result.

Google's guide

What do you need?

A Google Ads account

With admin rights, so you may create the connection.

A Google Cloud project

With the Google Ads API on and Explorer access.

An AI to work with

A chat app with MCP support, or a model API for an agent that runs by itself.

A measurement that is right

An agent that steers on wrong conversions does the wrong thing faster.

On your laptop or in the cloud?

On your laptopIn the cloud (Cloud Run)
Suited forAsking questions yourself in a chatAn agent that runs by itself and that your team shares
NeededPython, pipx, gcloud and your own Google loginCloud project, OAuth app and secrets in Secret Manager
Works when your laptop is offNoYes
CostYour AI subscriptionCloud Run and model usage, often a few tens of euros per month or less

Do you need the Gemini API?

Only if the agent runs by itself at fixed times, without anyone chatting. A worked example: a daily run with 200,000 tokens of data and 20,000 tokens of answer costs about $0.23 per day on Gemini 3.8 Flash, around $7 per month. That is the price through December 31, 2026, after which it doubles (Google). For account data, choose the paid tier: in the free tier Google uses your data to improve its products.

Google also ships ready-made skills for AI coding assistants (Agent Skills). Handy if you want to go further than the official MCP.

Where an agent pays off in Google Ads

These are the tasks where an agent in Google Ads pays off most right now. They come back all the time and a mistake can be caught. Search Engine Land names search terms as the first task to give an agent (Search Engine Land).

Going through search terms

Finding search terms that cost money without returning anything and excluding them.

In my crew: Pitch

Setting up campaigns

New campaigns, ad groups and keywords as a paused draft.

In my crew: Stack

Checking measurement

Checking that conversions keep coming in and that values are right.

In my crew: Tally

Writing the weekly report

What stood out and why, with the source for every number.

In my crew: Hoot

Making ad variants

Copy within your brand rules, with a check on price and promise.

In my crew: Leo

Recording changes

A log of who changed what and why, reversible.

In my crew: Hoot

How the crew works and which permissions each crew member has is on AI agents.

Four MCPs I build, with my way of working built in

Anyone can build an MCP these days. The difference is in the rules inside it. A generic MCP does what the AI thinks is good, and Google's standard recommendations often steer toward more reach and more spend. My MCPs contain my way of working from more than 12 years of Google Ads, at agencies and for large brands. So you do not get a tool, you get my way of working, looking along every day.

Fabiano as instructor with his four MCP robots lined up: Tally with an abacus, Stack with a wrench, a confused Hoot with his roll of paper and Leo with a paintbrush
Tally, the crew robot of this MCP

1Daily optimization check

Review MCP

In my crew: Tally

€2,000one-off

What it does

  • Goes through your search terms every day
  • Excludes clear waste by itself
  • Puts doubtful cases in the weekly report, approve with one click
  • Watches budget pace, disapprovals and measurement

My experience built in

  • When a search term is really waste, and when it is still too early
  • Which keyword type to exclude, so you do not block good searches
  • Brand protection and fixed exclusion lists per industry
Stack, the crew robot of this MCP

2Setting up campaigns

Build MCP

In my crew: Stack

€2,000one-off

What it does

  • Sets up campaigns, ad groups and keywords
  • Exclusions and extensions are in from day one
  • Everything starts paused as a draft
  • Going live only after your approval

My experience built in

  • My structure: brand, generic and competitors separate
  • Budget and bidding strategy that fit your margin, not maximum reach
  • Naming that keeps your reporting right later
Hoot, the crew robot of this MCP

3Shopping and Merchant Center

Feed check MCP

In my crew: Hoot

€2,000one-off

What it does

  • Checks disapprovals and warnings daily
  • Compares price and stock with your webshop
  • Finds products with weak titles or missing attributes, such as GTIN and brand
  • Prepares improvements and flags what needs attention

My experience built in

  • How a product title should be built to sell
  • Which attributes make the most difference in Shopping
  • Experience with feeds of thousands of products, as at Foto Semeins and Foto-Groep
Leo, the crew robot of this MCP

4Ad copy

RSA MCP

In my crew: Leo

€2,000one-off

What it does

  • Writes headlines and descriptions for responsive search ads
  • Respects the character limits and pinned positions
  • Checks every promise against your real price and offer
  • Prepares new variants as drafts to test

My experience built in

  • Copy that sells without empty superlatives
  • Your brand rules and banned words are fixed
  • What I have learned over years of testing about headlines that get clicks

All four together

The complete Google Ads crew

Separately the four MCPs cost €8,000 together. Have them built in one go and you pay €5,950. They then share one connection, one log and one weekly report.

€8,000 €5,950

All MCPs run in your own Google Cloud project. You own the connection, the data and the log. If you stop, you keep everything.

Build it yourself, have it built or outsource

There are three routes. They differ in money, time and in who holds the knowledge. The amounts for outsourcing are my own rates.

Fabiano with Tally and Hoot next to three stacks of coins that show the cost of the three routes
Build it yourselfHave me build itOutsource
One-offYour own time€2,000 per MCP, all four €5,950Often setup costs
Per monthRunning costs, about €10 to €25, plus your timeMaintenance from €500, plus running costsA specialist like me from €2,000
Year 1, approximately€120 to €300 plus your timeFrom €8,000 with one MCPA specialist from €24,000
Knowledge in the agentWhat you knowMore than 12 years of Google Ads experienceSits with the agency, not in an agent
Who decidesYouYou, with my rules in the agentThe agency, in consultation

Do I already manage your Google Ads? Then my crew is part of your management fee and you do not have to buy anything separately. Having it built is meant for teams that do their Google Ads themselves.

Work it out for your own team

Hours back per year
Value of those hours
Agent cost year 1
Difference in year 1
Difference from year 2

An example with 46 working weeks per year. How much work an agent takes over differs per account: enter your own estimate. The value of better decisions, such as less waste, is not counted here.

What it costs to have your agent built

Fixed prices, agreed up front. The build happens in your own Google Cloud project. Start with one MCP and expand later, or have the complete crew built in one go.

MCPWhat it doesOne-off
Review MCPDaily search term check with exclusions, monitoring of budget, disapprovals and measurement€2,000
Build MCPSetting up campaigns as a paused draft, following my structure€2,000
Feed check MCPShopping feed and Merchant Center checked every day€2,000
RSA MCPAd copy within your brand rules, with a price check€2,000
All four togetherOne connection, one log and one weekly report€5,950 instead of €8,000
€2,000from, per MCP

One MCP. Start with the task that costs your team the most time.

Book a call

€5,950one-off, all four

The complete Google Ads crew. Review, build, feed check and ad copy, built together and matched to each other.

Request a proposal

Maintenance from €500 per month, depending on the project: Google regularly releases new API versions and retires old ones, and your rules grow with your account. The AI automation scan of €2,950 is credited against the build. You pay running costs directly to Google. Prices excl. VAT.

Risks, and how to cover them

An agent can do something wrong with great confidence, and in Google Ads that costs money right away. These are the four risks I take seriously.

Risk

  • Too many permissions: a wrong bid or a deleted campaign happens fast
  • Hidden instructions: an agent that reads web pages or email can run into commands that are not yours (tops the list of OWASP)
  • Made-up numbers: a language model writes convincingly, even when a number is wrong
  • Steering on the wrong goal: an agent does not know your margin by itself

How I cover it

  • Small permissions, budget and bids only after approval (as OpenAI also advises, OpenAI)
  • Agents that read from outside get no write permissions
  • Every number in a report is checked against the source
  • Margin and goals are fixed in the rules, not in the model's guess

Gartner predicts that more than 40% of agent projects will be canceled by the end of 2027 because of cost, unclear value or too little risk management (Gartner). That is no reason not to start, but it is a reason to start small.

Checklist before an agent enters your account

Check these six things before an agent enters your account.

The account is in your name

Then you keep everything if you stop, including the log.

Your measurement is right

An agent that steers on wrong conversions does the wrong thing faster.

Permissions are small

Read only where possible. Exclude by fixed rules, the rest as a proposal.

There is a budget cap

Daily budgets and limits that an agent cannot raise.

It is a trial

The agent first runs alongside regular work.

There is a log

You see what happened and why, in plain language.

The specialist's role in 2026

An agent takes over routine work, the specialist keeps strategy, context and control. The specialist knows what your margin is, what your brand promises and what you do not want to say. The specialist also chooses which Google automation steers toward your goal and which mainly follows the platform's goal, such as AI Max (Google) and Performance Max, where you have been able to see channels and search terms since 2025 (Google).

As a freelance Google Ads specialist I use agents for the routine work and decide myself what goes live. Not sure which route fits? Start with the AI automation scan.

Frequently asked questions

What is a Google Ads AI agent?

A Google Ads AI agent is a system that works through a goal in your account by itself: fetching data, analyzing it and making or proposing changes. It uses a language model, tools and a check on its own result, within the permissions you give.

What is a Google Ads MCP?

MCP stands for Model Context Protocol, an open standard that lets an AI reach a system. The Google Ads MCP is the connection between an AI and your Google Ads account. Google has an official version that reads and creates assets and labels. A custom MCP on the Google Ads API can do more, such as adding exclusions and setting up campaigns.

Can an AI agent add exclusions by itself?

Not with Google's official MCP. With a custom MCP, yes. The review MCP I build excludes clear waste by itself under fixed rules and puts doubtful cases to you. Every change is in a log and can be undone.

Can an AI agent set up campaigns in Google Ads?

Yes, with a custom MCP on the Google Ads API. The build MCP sets up campaigns, ad groups, keywords and ads as a paused draft. Going live happens only after your approval. For keyword ideas from the Keyword Planner you need Basic access to the API.

Do I need the Gemini API?

Only if the agent runs by itself at fixed times. If you want to ask questions yourself, you connect the MCP to a chat app such as Claude, ChatGPT or Gemini CLI. For a scheduled agent, Gemini 3.8 Flash costs about $7 per month in a worked example, at the price through 2026.

Can a Google Ads agent run in the cloud?

Yes. The official MCP server runs on your laptop or as a service on Google Cloud Run, with login through OAuth. In the cloud the agent also works when your laptop is off, and your team can share it.

Does an AI agent replace the Google Ads specialist?

No. The agent takes over repetitive work, the specialist keeps strategy, context and control. That is why I build my experience into the agent's own rules.

Is a Google Ads AI agent safe?

Only with clear limits: small permissions, a budget cap, a trial run, a log and a human who approves budget, bids and going live. That is how I work myself, too.

What does it cost to have a Google Ads agent built?

Each MCP costs €2,000 one-off: the review MCP, build MCP, feed check MCP and RSA MCP. All four together cost €5,950 instead of €8,000. Maintenance is from €500 per month. The AI automation scan of €2,950 is credited against the build. If I manage your Google Ads, my crew is part of the management fee.

Need help?

Tell me what you run into.

Together we choose what you automate and what stays with a human.

Send a message

Start scanWhatsAppMail