Got something worth sharing? We want to hear from you.Share Your Knowledge →

Google Ads MCP Server Update: 4 New Write Tools in 2026

Listening to ArticlePlaying
0%
Google Ads MCP server update showing new AI write tools

Google’s Ads MCP server just picked up four write tools in version 0.0.6, moving it from a read-only reporting bridge to a tool that can touch live accounts. If you run paid search for a client or an in-house team and you have been testing AI agents against your Google Ads account, this update changes what those agents can actually do, and what they can accidentally do. This breakdown looks at what shipped, what still does not work, and whether the Google Ads MCP server is worth plugging into your stack right now.

What Is the Google Ads MCP Server

The Google Ads MCP server is an open source bridge that lets AI assistants such as Claude or custom agents talk to a live Google Ads account using the Model Context Protocol, the open standard that lets large language models call external tools in a structured way. Google’s Ads API team, led by engineer Dean Lukies, introduced it on October 7, 2025, and described the first release plainly: “This initial release is read-only, meaning it can be used for reporting and diagnostics,” according to the original announcement covered by PPC Land.

That first version shipped with two tools, search and list_accessible_customers, which let an agent pull campaign data and list the accounts it could see. Nothing could be changed. For agencies running fifty or a hundred client accounts, that alone was useful: an analyst could ask an agent to pull spend trends or flag budget pacing issues without writing a single GAQL query by hand.

What Changed in the Google Ads MCP Server Update

Version 0.0.6 shipped on PyPI on November 2, 2026, and it is a different kind of release. According to Google’s own developer integration guide, last updated November 11, 2026, the server is now described as “read and partial mutate”, not read-only. That is a meaningful change in scope for anyone who assumed the server could only look, not touch.

Four new tools handle the writing:

  • create_asset creates standalone assets, including text, images, YouTube videos, sitelinks, callouts, and structured snippets.
  • upload_video_asset uploads a local video file to YouTube through the Google Ads API as a background task, but only when the server runs in local stdio mode on the same machine as the client.
  • manage_labels creates, updates, and removes account-level labels.
  • manage_label_associations attaches or removes labels on campaigns, ad groups, customers, and ad group criteria such as keywords, though it does not support updating an existing association, only creating or removing one.

The practical gap worth flagging here, confirmed independently by Swiss marketing consultant Thomas Wismer in testing reported by PPC Land, is that assets created through create_asset land in the asset library and nothing else. “Assets only land in the library; I found no tool for the link in v0.0.6,” Wismer wrote, meaning an agent can generate a sitelink or a callout but cannot attach it to a campaign or ad group. You still need the Google Ads UI, the full API, or Google Ads Editor to finish the job.

The Propose-and-Confirm Workflow Behind Every Write

Every mutate tool in the Google Ads MCP server follows a two-step pattern. The first call defaults to validate_only set to true, which previews the change and returns a checksum instead of applying anything. A second call, sent with the same parameters and that checksum, commits the change.

ADVERTISEMENT

This sounds like a solid safety net, and it catches one real problem: it proves the parameters did not drift between the preview and the commit. It does not prove a human looked at the preview first. Wismer put it directly: “Whether a human approved it in between, the server cannot check.” Google’s own guide agrees, calling the checksum protection against accidental parameter drift, not an access control. Whether your agent pauses for your sign-off is a prompt instruction you write, not something the server enforces.

Where the Google Ads MCP Server Update Delivers

I would recommend testing this update if your team already uses AI agents for reporting and wants to extend that into light account maintenance, not full campaign management. A few things genuinely work well here.

  • Label management is the standout. Wismer’s test cycle, creating a label, reading it back, and removing it, worked cleanly, and labels are a low-risk way to let an agent organize accounts by tagging campaigns for reporting or scripts.
  • Asset creation saves real typing. Generating a batch of callouts or structured snippets from a brief, even if you still link them manually, cuts out a repetitive step agencies do by hand every week.
  • The confirmation step is visible, not hidden. Unlike a tool that silently applies changes, the checksum flow forces a second explicit call, which at least creates a clear point where a review step can sit.
  • It is free and open source. There is no license fee on top of your existing Google Ads API access, and the code is auditable on GitHub rather than sitting behind a vendor’s closed system.

Where the Google Ads MCP Server Falls Short

The honest version of this review has to sit with the gaps, because several of them matter more than the new features.

The biggest issue is a mismatch between documentation and default behavior. Google’s integration guide text in places still reads like the server is strictly read-only, while the shipped tools_config.yaml enables the write namespaces, labels and assets, by default. A team that installs the server by following the written setup steps without checking the config file can end up with write access it did not intend to grant. If you want a read-only deployment, you have to explicitly set labels: false and assets: false yourself.

A second problem is version drift. Google’s local setup instructions historically pulled from the main branch rather than a pinned release, so a server installed months ago can gain new write tools on its next restart with no changelog check required. Wismer’s recommended fix, pinning to a specific release tag such as 0.0.6, is the only real control here, and it is not the default.

Third, tool names changed. Starting with namespacing introduced in version 0.0.2 on September 9, 2026, tools carry prefixes like search_search or assets_create_asset instead of bare names. Any permission list or allow-list written against the old names can silently stop working. Wismer noted this is easy to miss: “I only noticed because I went through the release history.”

Fourth, remote deployment has a sharper edge now than before. If you deploy the server on Cloud Run with --allow-unauthenticated and skip setting an OAuth client ID and secret, the server falls back to Application Default Credentials, which can expose the container’s own service account to anyone who finds the endpoint. Before this update, that misconfiguration leaked data. Now it can also expose write access to account changes.

ADVERTISEMENT

Finally, scope creep is a real risk with labels specifically. Labels often drive automated rules, scripts, and reporting dashboards elsewhere in an account. An agent that relabels campaigns to “clean up” an account, acting on a vague instruction, can quietly break whatever logic downstream depends on those labels matching what they used to mean.

Who the Google Ads MCP Server Is For (and Who Should Wait)

This fits agencies and in-house PPC teams that already run AI agents for reporting and want to extend them into narrow, reversible account tasks, such as tagging campaigns by theme or batch-generating ad assets for a human to review and attach. It also fits developers comfortable reading a tools_config.yaml file and pinning dependency versions, because the safe defaults are not the shipped defaults.

It is a poor fit for teams that want a finished automation product rather than a developer toolkit, anyone managing budgets or bids through AI (the server still cannot touch either), and any account where “connect an AI agent” would happen without someone reviewing the tools_config.yaml file first. If your plan is full autonomous campaign management, this is not that tool, and nothing in Google’s current roadmap language suggests it is heading there soon. For that kind of broader AI campaign workflow, it is worth comparing how Meta’s own Ads MCP server handles the same read-versus-write tradeoff, since the two platforms have made different calls on default access.

Google Ads MCP Server: Read Tools vs. Write Tools in v0.0.6

The table below lays out exactly what each tool in the current release can do, and whether it needs the propose-and-confirm checksum step. Use it as a quick reference before you decide which namespaces to enable.

Tool Type What It Does Needs Checksum Confirmation
search Read Runs structured GAQL queries for reporting and diagnostics No
list_accessible_customers Read Lists customer IDs the authenticated user can access No
get_resource_metadata Read Returns selectable, filterable, and sortable fields for a resource type No
create_asset Write Creates text, image, video, sitelink, callout, or structured snippet assets Yes
upload_video_asset Write Uploads a local video to YouTube via the Ads API (local mode only) Yes
manage_labels Write Creates, updates, and removes account-level labels Yes
manage_label_associations Write Creates or removes labels on campaigns, ad groups, and keywords Yes

Caption: Seven tools make up the Google Ads MCP server as of version 0.0.6. The three read tools carry no risk beyond data exposure; the four write tools all route through the propose-and-confirm checksum flow, which guards against parameter drift but not against an agent acting without real human review. Data compiled from Google’s developer integration guide and PPC Land’s reporting.

FAQ

Does the Google Ads MCP server cost anything to use?
No. It is open source and free on GitHub and PyPI. You still need standard Google Ads API access and an approved access level for production accounts, which Google grants separately from the MCP server itself.

Can the Google Ads MCP server change my budgets or bids?
No, not in version 0.0.6. It cannot create or edit campaigns, ad groups, or ads, change budgets or bidding strategies, or add keywords. Those remain off limits through this server as of this update.

Is it safe to connect an AI agent directly to a live Google Ads account with this server?
It depends entirely on your configuration. The write namespaces are enabled by default, so connecting it without reviewing tools_config.yaml and pinning a release version carries real risk, particularly for label changes that feed other automation.

ADVERTISEMENT

The Verdict on the Google Ads MCP Server

The Google Ads MCP server update is a sensible, incremental step, not a leap into autonomous ad management, and that is mostly a good thing. Label management and asset creation are genuinely useful, low-stakes places to let an agent act, and the propose-and-confirm pattern is a reasonable design even if it leans on prompt discipline rather than server-enforced rules.

What keeps this from an easy recommendation is the gap between what the documentation implies and what the default configuration actually allows. A server marketed in places as read-only now writes to accounts out of the box, and the burden of locking that down sits entirely with whoever installs it. If you manage Google Ads accounts and want to experiment with this, pin the version, read tools_config.yaml line by line, and disable any namespace you are not actively using. Treat the checksum step as a nice-to-have, not a guardrail, and write your own approval gate into the agent’s instructions. Used that way, with eyes open about what it cannot do and who can misuse what it can, the Google Ads MCP server earns a place in a PPC team’s toolkit well ahead of most AI-agent integrations that promise far more and document far less. For a broader look at how these integrations fit together, our guide on how AI agents are actually working in marketing teams covers where agents are earning their keep beyond ads platforms, and our walkthroughs of Rank Math’s MCP tools and Surfer’s MCP update show the same pattern playing out on the SEO side of the stack.

Responses

Leave a response

Your email address will not be published. Responses are reviewed before they appear.

Join the DevByte Daily list

AI, SEO and paid media coverage in your inbox. No spam.

ADVERTISEMENT

Share This

Questions or story tips?

Contact us

Enjoyed this article?

AI, SEO and paid media coverage in your inbox. No spam.