Google Ads MCP write access: what AI can change in your account

Yes, Google's official Google Ads MCP server can now make changes, but only small ones. Version 0.0.6, released on November 2, 2026, lets an AI assistant create assets, upload videos and manage labels after a dry run, and the write tools are switched on by default. It still cannot touch campaigns, ad groups, ads, budgets, bids or keywords, and none of the new tools reach Merchant Center, where most Google Shopping results are decided.

This guide is for e-commerce teams and agencies deciding how much an AI assistant should be allowed to change. It covers what the server can and cannot do as of November 2026, how the confirmation step works, a naming trap around labels, a safety checklist before you allow writes, and what this shift means for the way PPC agencies work.

What changed in Google's Google Ads MCP server?

MCP (Model Context Protocol) is the standard that lets an assistant such as Claude or ChatGPT call outside tools. Google publishes an open-source MCP server for the Google Ads API on GitHub. Until late October 2026 it could only read: run Google Ads Query Language (GAQL) searches, list the accounts you can access, and look up field metadata. Our guide to connecting Claude to Google Ads with this MCP server covers that read-only setup for online stores.

The v0.0.6 release notes add one line that changes that: "Implement asset and label management with propose-and-confirm flow." The pull request behind it, merged on November 2, 2026, describes it as "initial support for mutate capabilities" and states that the mutate tools are enabled by default in the server's tools_config.yaml file.

Google has not announced the change in a blog post, and as of November 10, 2026 the Google Ads MCP server developer guide still lists the mode as "Read-only (current release)". If you installed the server earlier and update it, you get the write tools without opting in.

What can the Google Ads MCP server change, and what can't it?

The table below is based on Google's release and pull request, and on a hands-on test published by Thomas Wismer on November 9, 2026 (Wismer Marketing), which lists the tool names. Check the server's README for your version before relying on it.

In practice, the write side today is housekeeping: building an asset library and organizing the account with labels. That is useful, but it is not where a Shopping account wins or loses money.

How does propose-and-confirm work, and does it prove a human approved?

Every write runs in two steps. First the server sends the change to the Google Ads API in validate-only mode, which checks it without applying it, and returns the planned change. Then a second call commits it. According to Wismer's test, the first call also returns a checksum, and the commit is rejected if its parameters do not match the proposal.

That protects you from an assistant committing something different from what it proposed. It does not prove a person read the proposal. The server cannot tell whether you approved or the assistant simply made the second call itself. Wismer adds his own approval step in his AI client so that he is asked before every commit. If your client runs tools automatically, the confirmation step alone is not a human review.

One more safety net exists outside the MCP server. Google Ads change history lists account changes for the past 2 years, and Google says changes made through the API can show "Google Ads API" or the tool name in the user column. Check it after any session where an assistant was allowed to write.

Are Google Ads labels the same as Merchant Center custom labels?

No, and this is the trap for Shopping advertisers. The labels the MCP server can now manage are Google Ads labels: tags you put on campaigns, ad groups, ads and keywords to filter and report on them. Google notes they are not inherited, so a label on a campaign does not apply to its ad groups.

Merchant Center custom labels are product attributes, custom_label_0 to custom_label_4, that you send in your product data. Google says they let you build filters in Performance Max (PMax), Shopping and Demand Gen campaigns for reporting and bidding on groups of products. Each of the five can hold up to 1,000 unique values across the account, and values beyond that are ignored for reporting and bidding.

So when someone says "the AI can manage labels now", it does not mean it can move a product from a high-priority group to a low-priority one. That still happens in the feed. Clarmix's Product Labelizer works on that side: it labels each SKU daily as Profitable, High Potential, Low Potential, Costly, Zero Conversion or Zombies from clicks, target ROAS and cost per conversion, and sends the labels to Merchant Center through a supplemental feed so the primary feed is untouched. For the basics, see our guide on what a labelizer is and how custom labels work.

Why are most Shopping levers outside the Google Ads MCP server?

In a Search account, many decisions live in Google Ads: keywords, ad copy, bids. In Google Shopping and PMax, a large share of them live in the product data. Titles and product types decide which searches a product can match, custom labels decide how products are grouped and bid, and Merchant Center issues decide whether a product shows at all.

None of that is reachable through the Google Ads API tools. An assistant with write access to Google Ads can put a label on a campaign, but it cannot fix a disapproved best seller or a vague title. Feed changes still go through Merchant Center, a feed tool or a supplemental feed. Clarmix's feed enrichment keeps a person in that loop: AI rewrites of titles, descriptions, Google product categories and product types are reviewed and approved by the user, then sent through a supplemental feed.

If you are segmenting a Shopify catalog, the same logic applies to margin data. Our guide to segmenting Shopify products by margin shows how profit data becomes custom labels and campaign structure.

How should an agency set up write access safely?

If you run the Google Ads MCP server for clients, decide on write access per account rather than inheriting the default. A short checklist:

1. Decide if you need writes at all. If the job is analysis and reporting, turn the write tools off. The README explains that the server reads its tool settings from a tools_config.yaml file, which you can point to with the GOOGLE_ADS_MCP_TOOLS_CONFIG environment variable; Wismer disables the asset and label groups there.
2. Pin the version. A routine update gave existing installs write tools. Pin the package version and read the release notes before upgrading.
3. Add a real approval step. Configure your AI client to ask before any committing call, instead of relying on the server's confirmation flow.
4. Scope by account. Connect with credentials that only reach the accounts the assistant should touch, and be explicit about the customer ID in every request.
5. Separate read and write setups. Keep a read-only configuration for day-to-day analysis and a separate one for the occasional write session.
6. Review change history. After each write session, check the Google Ads change history and note what the assistant changed in your client log.
7. Tell the client. Agree in writing which kinds of changes an assistant may make and who approves them.

Where does the Clarmix MCP server fit?

The Clarmix MCP server takes the opposite position on writes. Clarmix describes it as read-only: its 26 tools cannot change anything in Clarmix, Google Ads, Merchant Center or Shopify. What it adds is reach. One connection in Claude, ChatGPT or another MCP client reads Google Ads, Merchant Center, Clarmix labels and audits, and Shopify data through the Clarmix Shopify app, so the assistant can answer questions such as which disapproved products were top spenders or which products spent the most without a conversion.

We think that split is the sensible one for most stores today: let the assistant see everything and diagnose across sources, then make changes in the tool built for that change, with a person reviewing it. Our guides on building a Google Shopping AI agent in Claude and ChatGPT and on Shopify MCP servers for Google Ads cover the setup.

What is the future of PPC agencies if AI can edit accounts?

Write access in an official Google tool brings the question closer for many agencies: should account managers still make changes by hand in the Google Ads interface, or move to strategy and leave the edits to AI?

At Clarmix we think "strategy only" is the wrong answer. Advice from someone who no longer knows the account well is weak advice, and someone still has to approve changes, debug a drop in results and handle everything the tools cannot change yet, which today includes campaigns, budgets, bids and keywords.

What changes is the hands-on work. Less of it is clicking through routine edits and building reports, which assistants already do quickly. More of it is setting the rules the automation works within:

• targets and tolerances based on real margins, not revenue alone
• product segments and custom labels that decide where budget goes
• feed quality and Merchant Center health
• exclusions, brand rules and what an assistant may change without asking
• checking what Google's automation and any AI assistant actually did

The case for the other side is real: for small accounts with simple catalogs, an agency that only sets direction and lets automation run may be enough, and some clients will prefer to pay less for that. For stores with large catalogs and thin margins, we expect clients to keep paying for judgment, for someone who answers for the results, and for the data a generic assistant does not have: margins, stock and product feed health.

Frequently asked questions

Is the Google Ads MCP server read-only?

Not any more. Version 0.0.6, released November 2, 2026, adds asset creation, video upload and label management with a propose-and-confirm flow. Google's developer guide still described the server as read-only on November 10, 2026, so check the release notes for the version you run.

Can an AI assistant change my bids or budgets through the Google Ads MCP server?

Not as of November 2026. The write tools cover assets and labels only. Campaigns, ad groups, ads, budgets, bids, status and keywords are not in v0.0.6.

Can the Google Ads MCP server set custom labels for Shopping?

No. It manages Google Ads labels, which organize campaigns, ad groups, ads and keywords. Custom labels for Shopping and PMax are Merchant Center product attributes and are changed in your product data.

How do I turn off write access?

The server reads its tool settings from a tools_config.yaml file, which you can point to with the GOOGLE_ADS_MCP_TOOLS_CONFIG environment variable. Disable the asset and label tool groups there, then restart the server.

Can the Clarmix MCP server make changes?

No. Clarmix describes it as read-only across Clarmix, Google Ads, Merchant Center and Shopify. Changes stay in those tools or in the Clarmix app, where enrichments are reviewed before they are sent.

Sources

• Release v0.0.6, googleads/google-ads-mcp (GitHub)
• Implement asset and label management with propose-and-confirm flow, pull request #139 (GitHub)
• googleads/google-ads-mcp README (GitHub)
• Google Ads MCP server: Developer integration guide (Google for Developers)
• About ads labels (Google Ads Help)
• Custom label 0–4 [custom_label_0–4] (Google Merchant Center Help)
• About change history (Google Ads Help)
• Google's official Google Ads MCP server can now write (v0.0.6), Thomas Wismer, Nov 9, 2026 (Wismer Marketing)
• Clarmix MCP server
• Clarmix Product Labelizer
• Clarmix feed enrichment

Cover photo by Cash Macanaya on Unsplash.