Can the Official Google Ads MCP Create Campaigns, or Is It Read-Only?

Can the Google Ads MCP create campaigns?

No. Google's official Google Ads MCP is read-only. It ships with exactly three tools — list_accessible_customers, search, and get_resource_metadata — and every one of them is a read. It can pull your campaigns, break down your Quality Scores, and reason about a CPA spike in plain English. It cannot create a campaign, pause an ad, change a bid, or edit a single line of ad copy. That is a deliberate choice by Google, not a gap they forgot to fill.

If you have been told an AI assistant can now "run your Google Ads for you," this is the part nobody mentions: the connector Google actually publishes only lets Claude look. Writing to the account is a separate, gated path — and it is worth understanding before you go hunting for a workaround.

First, you have to connect (and host) the Google Ads MCP

Before any of this matters, you need the MCP connected — and Google makes that harder than Meta does. Meta hands you a hosted connector you just sign into; Google ships a repository you host yourself, so there is no one-click link to paste. Standing it up (on Cloud Run, Docker, or locally) is a one-time job, and you also need a manager (MCC) account and a developer token. We walk through exactly why there is no simple link and how to host it in the Google Ads MCP: what it can and can't do. Get it connected first; then the read-only limit below is what you are working within.

Why Google made it read-only (and Meta didn't)

The philosophical split between the two platforms

Meta and Google took opposite stances on AI write access. Meta hands you a hosted connector that will create and edit campaigns conversationally. Google ships a read-only server that you host yourself, and routes every write through its existing API and approval process. Same idea — "let an AI talk to your ad account" — two very different levels of trust.

We get the same question in almost every webinar, close to verbatim: "Wanted to know if this Google MCP, like we can use it to analyze the data like you just showed? Or can it also be used to create ads or maybe optimize the ad copies, ad creatives?" The honest answer on camera: "Unlike Meta, Google has kept it read-only. You cannot actually make changes using that. So the step forward in that is to use the APIs directly. But Google will have their approval process — you will have to submit the PRD."

What read-only actually buys you — and it isn't nothing

Read-only sounds like a limitation until you have watched an AI confidently push a change you never sanctioned. A read-only connector cannot burn budget, cannot pause the wrong campaign, and cannot "optimize" your bids into a wall at 2am. It analyses; you decide; you execute. For most teams putting AI anywhere near a live ad account, that is exactly the right blast radius — and the reason the read-only limit is less of a problem than it first looks.

So what can you do if you want to create campaigns?

Plenty — you just don't do it through the official MCP. There are three real paths, in rough order of effort and risk.

Option 1 — the Google Ads API directly, and the approval process

The write path Google intends is the Google Ads API, not the MCP. You apply for a developer token, and for anything beyond your own test accounts you submit for review — Google wants to see what you are building and how it will touch accounts (the "PRD" step people trip over). It is the most powerful option and the most work: real engineering, a real approval wait, and ongoing maintenance. Right for a team building a product or a standing automation; overkill if you just want to launch a few campaigns faster.

Option 2 — a community MCP with write access

Community-built MCP servers exist that wrap the write API, so Claude can create and edit through them. They can genuinely work — but you are handing a third-party server the keys to a live ad account, so vet it like one. Check the fork and star counts, read the code that touches authentication, and be wary of a repo with very few users. Very few users is usually a danger signal, not an early-adopter badge. If you cannot read the auth code and satisfy yourself it is safe, don't connect it to a spending account.

Option 3 — Claude drafts, you execute

The underrated one. Let the read-only MCP do the thinking — read the account, find the waste, and draft the new campaign structure, budgets, and ad copy — then you paste the result into Google Ads yourself. It keeps the AI where it is strongest (analysis and drafting) and keeps your hand on the actual write. No API approval, no third-party server holding your credentials, no 2am surprises. For most marketers evaluating AI on a real account, this is the fastest safe path to "campaigns created with Claude."

Option	Effort	Risk	Who it's for
Google Ads API directly	High — dev token + approval + build	Low, once approved	Teams building a product or standing automation
Community MCP with write access	Medium — find and vet a repo	High — third party holds live credentials	Technical users who can audit the code
Claude drafts, you execute	Low — works today	Minimal — you make every write	Most marketers who want speed, not autonomy

What we'd actually recommend

Why read-only is the right phase one for most teams

Across account after account, the operators we work with independently draw the same line: automate the analysis, keep hands on the execution. Read-only is not a stage to rush past — for a first deployment it is the correct phase. Let Claude read the account, run the audit, and draft the changes; keep the write for a human until you trust the pattern. When you genuinely outgrow it, Option 1 is there. Most teams find they do not need to.

This is exactly the stack we run Google Ads audits on: the read-only MCP for analysis, our own framework for judgement, and a human on the write. If you want that run against your account, that's what our Google Ads audit does.