A Google Ads MCP server lets an AI assistant run live, structured queries against a Google Ads account, so questions about performance, structure, or configuration get answered from the account as it exists right now instead of from an export that froze at download.
Akorn's Google Ads connector is read-only and runs in production across client accounts. It answers performance questions (spend, conversions, ROAS by campaign, ad group, or asset), structural questions (what exists, what changed, when), and configuration questions (which conversion actions are wired to which campaigns). It cannot change anything in an account. Its job is verification, not automation.
Three kinds, in practice:
Performance. Which campaign had the best return this month. What the search terms actually were. How this week compares to the same week last year. The routine reporting layer, minus the export step.
Structure and history. What exists in the account, and what changed. Change history queries answer "who touched this and when" without reconstructing it from memory.
Configuration. The quiet one, and the most valuable. Whether a conversion action is actually attached to a campaign's goal. Whether a bid strategy is what the plan says it is. Whether the thing marked done in the project tracker exists in the account.
That last category is where the catches live.
Conversions inflated 6.5x. On one account, the headline conversions metric was silently summing every conversion action together: purchases, sign-ups, everything. Reported conversions ran 6.5x above reality. No dashboard view flags this, because the number is technically correct; it is just answering a different question than anyone thinks they are asking. A query segmented by conversion action name showed the blend immediately. The fix was a reporting rule, not a campaign change, and every number since has described what it claims to describe.
"Done" that was not done. A checklist recorded video conversion goals as configured on June 26. A live pull on July 1 found zero such conversion actions anywhere in the account, while the campaigns that depended on them were already spending. The task had been assigned and marked complete; the account said otherwise. Verification against the live system is the only thing that distinguishes those two claims.
A metric that changes meaning by campaign type. Cost-per-acquisition on Performance Max has to be read from a different conversions column than Search. Read the wrong one and PMax cost per booking looks roughly 3x better than it is. This is not a bug; it is documented API behavior that almost nobody operationalizes. Our standing rule specifies the column per campaign type, so the number cannot drift based on who ran the query.
The connector alone is not the system. The rules on top of it are:
Filter by campaign ID or channel type, never by name, because names change and name-based queries fail silently. Read the conversions column appropriate to the campaign type, every time. Run the conversion-action sanity check before any analysis, and halt if the account contains actions the ruleset does not expect. Verify any figure headed into client communication against the platform interface, because a live query is a tool, not an oracle.
The result is that the same question produces the same correct answer on any day, run by anyone.
It cannot pause, edit, bid, or spend. It does not see a handful of UI-only surfaces, such as auction insights competitor rows, which still get captured manually and merged. And it answers "what is true now" better than "what was true daily for the past year"; deep historical trending needs data that was deliberately archived along the way.
Is read-only actually enforced, or just promised?
Enforced. The query layer is select-only; there is no code path that constructs a change. Every query is also logged against the person who ran it.
Does this replace a PPC manager?
No. It replaces the export-and-reconcile step. Deciding what a discrepancy means, and what to do about it, is still the job.
Can it audit an account we do not manage yet?
Yes, with read access granted. A structured read of conversion configuration, query routing, and change history is the backbone of how Akorn evaluates an account before touching it. Book the AI-Stack Evaluation
Why not just use the Google Ads UI?
The UI answers one question per screen. A query answers a question across the whole account at once, and cross-references it against Merchant Center, Tag Manager, or Analytics in the same conversation. Read: the live-query stack hub →
A structured read of conversion configuration, query routing, and change history — the backbone of how Akorn evaluates an account.
Get in touch