An MCP server for Google Ads lets an AI assistant query an ad account directly through the Google Ads API, so performance and configuration questions get answered live instead of through exports. Several real options exist, including one from Google itself. This page inventories them honestly, including the free ones, because the useful comparison is not "which server" but "read-only or write-capable, self-hosted or operated, and who owns the rules on top."
Google open-sourced an official Google Ads API MCP server in October 2025; its initial version is read-only and built for diagnostics and analytics, and for a single-platform read it is genuinely good and free. Community open-source servers exist with wider scope, several of them write-capable, meaning they can pause campaigns or change budgets. Hosted agent products layer automation and interfaces on top for a subscription. Akorn's fleet uses the same protocol; the difference it sells is not access, it is cross-system operation with a maintained rulebook and the catches to show for it.
Google's own server (googleads/google-ads-mcp on GitHub) shipped publicly in October 2025 and is documented in the Google Ads developer toolkit. The initial version is deliberately read-only, aimed at diagnostics and analytics. It runs locally or on your own cloud, and setup is a real developer task: a Google Ads API developer token with production access, OAuth credentials, and hosting are all on you.
Where it fits: a technical team that wants a trustworthy, free, single-platform read into Google Ads, and is comfortable owning the setup and the credential handling. For that job, start here.
The community landscape around the google-ads-mcp topic on GitHub is active, and the entries differ on the axis that matters most: whether they can change things.
Analysis-focused, read-oriented options exist (cohnen/mcp-google-ads is a widely referenced early one, MIT-licensed, built for conversational analysis in Claude and similar clients). Broader projects go the other way on purpose: some implement large tool surfaces including campaign management, and at least two verified this week expose write operations such as pausing campaigns, updating budgets, or creating campaigns outright. Multi-platform servers also exist, spanning Google Ads, Meta, GA4, TikTok, and LinkedIn in one package, and bundles covering several Google services together.
Three cautions that apply to all of them. First, decide deliberately whether you want write capability to exist at all; an AI that can move budgets is a different risk class than one that can only look, and the default in this house is that it should only look. Second, check maintenance before adopting anything: last commit recency, open issues, API version support. Repos in this space go stale fast, and the Google Ads API deprecates versions on a schedule. Third, review how a project handles credentials before you paste yours anywhere; treat any example configuration containing real-looking values as a red flag about the project's hygiene.
A separate category sells the outcome rather than the server: subscription products (Adsroid and Soku are two currently marketing in this space) that connect to ad accounts and layer agents, alerts, and reporting on top. The trade is the usual one: less setup and a product interface, in exchange for a vendor in the loop and their defaults instead of yours. Evaluate them on the same axes: what can it write, what does it log, and whose rules decide what a number means.
None of these, official or otherwise, comes with the operating rules that make the answers reliable: which conversions column is correct for which campaign type, how to keep queries valid through campaign renames, what sanity check runs before an analysis is trusted. The connection is the commodity half. The rulebook, and the discipline of applying it every time, is the half that determines whether the fast answer is also the right one. Read: the live-query stack hub →
Same protocol as everything above. Five services operated in production (Google Ads, Merchant Center, Tag Manager, GA4, Search Console), read-only by policy across all of them, every query logged, run for clients as part of an engagement rather than handed over as software. If your need is a single-platform read and you have a technical team, Google's free server is the honest recommendation. The case for an operated fleet starts where the questions cross systems and the answers have to be right every week regardless of who asks: that is a service claim, and the catches it has produced are documented across this site. Read: the Google Ads MCP → Read: 7 things live queries catch →
Is there an official Google MCP server for Google Ads?
Yes. Google open-sourced one in October 2025, read-only in its initial version, available on GitHub and documented in the Google Ads developer toolkit.
Are the open-source options safe to connect to a live account?
The official server's read-only design is the safer default. For community servers, check what the tools can do before connecting: several can pause campaigns or change budgets, which you should treat as a deliberate decision, not a footnote.
Do I need any of this if I already have dashboards?
Different jobs. A dashboard is a scheduled summary; a live query answers the questions the dashboard was not built for. The longer comparison, including where data pipelines fit: Read: MCP vs pipeline vs dashboard →.
What does it cost?
The official and community servers are free software plus your setup time, hosting, and API access. Hosted products are subscriptions. An operated fleet is part of a service engagement, which is a different purchase for a different problem.
Entries verified September 3, 2026. This page is re-verified quarterly; the ecosystem moves fast enough that anything older than a quarter should be re-checked before you rely on it.