AI and MCP tools for Google Ads in 2026: the official server, open source and hosted connectors

A year ago, connecting Google Ads to a language model meant opening your own API project. Today there are three routes: Google's official MCP server, open-source servers built by the community, and hosted connectors such as Tecrube. All three carry the name “MCP”, but what they can do differs widely.

The wrong choice costs you in one of two ways: either the tool cannot pause a campaign and you end up back in the dashboard, or it has write access with no approval mechanism and one misread sentence changes a budget.

We judged on five questions: read or write; does setup require code or an API project; is there a preview and approval on writes; which clients does it work with; does it cover platforms beyond Google.

Price is deliberately not a criterion, because open-source servers look free while setup and maintenance time is the real cost.

The official server grants read access to the Google Ads API and is reliable for pulling reports. It is read-only by design: it cannot pause a campaign, move budget or change a bid. Setup requires creating a developer token and an OAuth client.

It is the right pick if you only want reports and analysis, are comfortable with the technical setup, and will keep making changes in the dashboard.

There are several open-source Google Ads MCP servers on GitHub, some of which support writes. The advantage is transparency and flexibility: you can read the code and run it on your own server.

The downside is responsibility. When the API version changes the update is yours, credentials live on your machine, and most have no approval gate for writes. It makes sense for companies with a developer team that can write their own safety layer.

Tecrube is a hosted MCP connector added to Claude, ChatGPT, Claude Code and Cursor with a single address. There is no API project, token or server to set up; you link your Google account with OAuth. It covers both reads and writes.

On the write side every action follows preview → approve → apply, budget increases above fifty percent are rejected and every change is written to the audit log. Alongside Google Ads, Meta, Microsoft, TikTok and LinkedIn as well as GA4, Tag Manager and Search Console sit behind the same connection. The free account gives 75 credits that never expire; Pro is $49 a month and Agency $99.

Read only and no fear of technical setup: the official server. Full control on your own infrastructure with developer resources: open source. Reads and approved writes without code, across several platforms: a hosted connector.

Using more than one MCP at the same time is possible; you can add the official server for reports and Tecrube for changes to the same Claude session. Giving AI write access is not risky as long as the approval gate lives on the server; the risk is in not having a gate at all.

Whichever tool you pick, answer these five questions before setting it up. Will the assistant need to pause campaigns or change budgets, or do I only want reports? Where will the credentials live, on my machine or on a server? Can I see a write before it is applied?

Should my Meta, Microsoft or LinkedIn accounts sit in the same chat as Google Ads? And if several people on my team use the tool, how will I later see who changed what? The answers usually decide on their own which of the three options fits.

In Tecrube the answer to the last two is the audit log and a single connector: every change is recorded with account, user and time, and all five ad networks share one address. To try it, the 75 free credits cover a few audits and a change so you can see the flow on your own account.

Whichever tool you choose, working only with reads in the first week and seeing how it interprets the data is the healthiest step before trusting it with write access.