Google's free MCP server: what it does and what it doesn't
When Google announced an official MCP server for Google Ads, the reaction split neatly in two: marketers who assumed it would do everything, and engineers who read the scope document and understood it would do almost nothing that touched money. Both groups need a clearer picture of the actual boundary.
What the Google MCP server actually ships
Google's connector grants one scope: read. You can pull campaigns, ad groups, keyword performance, impression share, and budget pacing. That is a useful data surface — you can ask your AI client "which campaigns are under-pacing this week" and get a real answer without opening the Ads UI.
What it does not ship: any write scope. create_campaign, adjust_budget, pause_ad_group — none of those tools exist in the official connector. The model can observe but not act.
This is a deliberate product decision, not an oversight
Where the ceiling shows up in practice
Say you ask Claude: "The branded campaign is over-pacing by 40%. Reduce the daily budget to $180." With the Google connector, Claude can confirm the pacing, quote you the current budget, and describe what you should do. It cannot do the thing.
You close the conversation, open Ads Manager, find the campaign, click Edit, change the number, and save. The AI became a research assistant, not an operator. That is a meaningful difference in time-cost per account, especially if you manage more than five.
The draft/approve model fills the gap
SpendSignoff adds write capability under a hard safety contract: the model stages changes as before→after diffs, and a human approves in a two-step confirm before anything touches the account. The model never holds an mcp.approve scope — only mcp.read and mcp.draft.
The result is that the workflow above completes in the AI client. You get the pacing alert, review the staged $180 budget draft, click Approve, and the change is pushed. The approval step is yours; the draft creation is the model's.
Approval is server-enforced, not UI-enforced
When the free connector is the right choice
If your workflow is purely analytical — building reports, auditing account structure, feeding data to another system — the Google connector is fine. It is free, maintained by the platform owner, and reliably scoped to never surprise you with a spend change.
If you want a model that can act on what it finds, you need a server that ships write capability under a controlled approval gate. That is the gap SpendSignoff was built to fill.
FAQ
- Will Google ever add write access to its official MCP server?
- Possibly, but the timeline is unknown. Platform-level write access through third-party AI clients raises compliance and liability questions that take time to resolve. The safer assumption for 2026 planning is that the official connector stays read-only.
- Can I run both the Google connector and SpendSignoff at the same time?
- Yes. They are separate tools in your AI client's tool list. You can use the Google connector for data pulls that need no action, and SpendSignoff for operations that require a draft and approval.
- Does SpendSignoff compete with the Google connector, or complement it?
- Complement. SpendSignoff extends what the read layer enables by adding a safe write path. It does not replace the Google connector's read coverage.
Connect an account read-only and watch the operator work.
Reads are free on every plan. Nothing spends without your two-step approval.
Related reading