Google open-sourced a read-only Model Context Protocol (MCP) server for Google Ads that lets AI assistants securely query account data and return natural-language diagnostics. It cannot change campaigns, but it can accelerate audits and reporting for contractors by surfacing waste, schedule and geo issues, device splits, and competitor pressure. The post outlines practical workflows like weekly waste checks, call-quality comparisons, geo/schedule guardrails, seasonal budget pivots, and ad asset triage—all with humans making the final changes. It warns about attribution quality, least-privilege access, and building guardrails now to prepare for potential future write access.

Google’s New MCP Server for Ads: Real Ways Contractors Can Use It Today (Skip the Hype)

TL;DR Google open-sourced a read-only Ads API “MCP server” that lets AI tools pull diagnostics and insights via natural language. It won’t build campaigns (yet), but it can speed up analysis, spot waste, and surface what moves the phone. Treat it like a sharp analyst, not an autopilot.

What Google Actually Released (and Why You Should Care)

Google has open-sourced a Model Context Protocol (MCP) Server for Google Ads. In plain English: your AI assistant can securely “read” your Google Ads data and answer questions like a smart analyst. The initial release is read-only—no changes to campaigns—so think diagnostics, auditing, and reporting, not “set it and forget it.”

That’s not a bad thing. For HVAC, plumbing, and electrical shops, misfires usually hide in the boring parts: search terms, schedule bleed, service area waste, and calls routed to the wrong team. An AI layer that fetches and explains the right data on demand can cut hours and expose “why CPL spiked” before your dispatcher starts yelling.

What You Can Do Right Now (Read-Only, But Useful)

The MCP server lets a large language model (LLM) securely query your Ads account and answer in natural language. Examples of real questions you can ask and get supported by live data:

• Which search terms in the last 14 days spent over $200 with zero calls or form leads?
• Compare after-hours (6 pm–7 am) spend and conversion rate week over week. Is our emergency HVAC ad schedule paying off?
• Which zip codes have the worst CPL for drain cleaning, and are they outside our preferred service radius?
• Break down conversions by device for “AC repair” vs “furnace repair” ad groups. Where are we overpaying?
• Show the top auction competitors impacting impression share on “electrical panel upgrade.”

All of this is classic diagnostics—faster, cleaner, and less clicking. The LLM translates your question, calls the right metrics, and returns a narrative with supporting numbers. You still make the decision, but you’ll get there faster.

How I’d Use It for Contractors (Calls, Not Clicks)

Here’s how I’d plug MCP into a contractor workflow without adding fluff:

• Weekly “Waste Watch” brief: Find spend over threshold with zero conversions by campaign/ad group/search term. Flag negatives to test and hours to cut.
• Call-first conversion checks: If you import calls as conversions, have the AI compare total conversions vs qualified call outcomes. Surface campaigns with lots of short calls. You’ll find junk fast.
• Geo and schedule guardrails: Ask for heat maps by zip and hour. Trim where leads are low-intent or far from profitable routes.
• Seasonal pivots: Build a standing report that watches cooling vs. heating category performance and suggests budget shifts as weather changes.
• Asset triage: Pull headlines/descriptions with lowest CTR and highest CPC in search campaigns. Use it to prioritize ad tests that actually reduce CPA.

None of this requires the AI to “touch” your campaigns. It just gets you to the right lever faster so you can pull it.

Limits and Gotchas (Don’t Hand the Keys Over)

Read-only means safer, but not foolproof. LLMs can still be confidently wrong. Always spot-check the numbers inside Google Ads before acting.

• It can’t make changes—yet. Good. Keep a human in control of budgets, bids, and structure.
• Privacy and access: Grant least-privilege, read-only access. Keep customer PII out of prompts. Lock down which accounts the AI can see.
• Attribution muddiness: If your conversion tracking is sloppy, the AI will parrot bad data. Fix tracking first: calls, forms, revenue where possible.
• Over-automation risk: When write access arrives, build guardrails: change limits, approval queues, and rollback plans.

What’s Coming Next (and How to Prepare)

Google hints that future versions may go beyond read-only. That’s where people get into trouble if they haven’t built a process. Prep now:

• Create “approved playbooks” the AI can propose, not execute: pause waste by rule, schedule trims, negative keyword bundles, budget reallocation caps.
• Require human approval for changes over a set spend threshold or in peak season.
• Test in a sandbox or low-risk campaign first. Measure before/after CPL and call quality.

Quick Start Checklist

1. N/A