PPC.co

Google's official MCP server lets AI assistants query live Google Ads data directly — no custom integration required. This episode breaks down how it works, what it can't do, and how to wire it up correctly for real agency workflows.

Show Notes

AI-powered PPC reporting just got a lot more practical. Google's open-source Model Context Protocol server for the Google Ads API lets large language models query live account data directly — no middleware, no stale exports, no custom integration tax. This episode walks through exactly how the setup works, where the hard limits are, and how paid media teams can fold this into a real workflow today. The full technical walkthrough is available in the source article on connecting Google's MCP server to Google Ads.

Here's what the episode covers:

  • What MCP actually is: Anthropic's open standard gives large language models a consistent, shared way to connect to external data sources — Google adopted it for the Ads API, collapsing the need for one-off integrations.
  • Read-only by design: The official server ships with three tools — account listing, a GAQL query runner, and a metadata helper. No bid changes, no budget edits, no campaign pausing; that constraint is intentional and a meaningful safety feature.
  • The four credentials you need: A manager account with an approved developer token (scoped to the MCC), a Google Cloud project with the Ads API enabled, an OAuth 2.0 credentials file, and Python 3.12 or later — all four must be in place before anything works.
  • Which AI clients can connect: The server runs locally over stdio transport, meaning Claude Code, Cursor, Gemini CLI, and VS Code work natively — but web-based apps like Claude.ai and ChatGPT cannot reach a locally hosted process.
  • Guardrails worth building in: Scope OAuth credentials to only the accounts the model needs to see, log every tool call for a full audit trail, keep developer tokens out of shared chats, and filter query results server-side when volumes are high to control context-window costs.
  • Where it fits in agency work: The MCP layer functions as a frictionless reporting interface — diagnostic questions that once required spreadsheet pulls now resolve in a single prompt — but account strategy, keyword hygiene, and full-funnel thinking remain with the operator.

For more on managing PPC performance through challenging periods, check out the episode How to Beat PPC Seasonality Issues and Win the Slow Season.

PPC.co

What is PPC.co?

Paid search and paid social judged on profit rather than platform dashboards. Bidding structure, account architecture, creative testing, incrementality, audience strategy, and the budget conversations that decide whether a channel survives the quarter.

Every episode takes a single decision a paid media manager is weighing — consolidate campaigns or split them, trust the platform's attribution or not, kill a channel or fund it properly — and reasons it out loud. It's written for the person who has to defend the spend, not the person who read a case study about it. Five or six minutes an episode. No guest interviews, no vendor pitches.

Topics include account structure and campaign consolidation, smart bidding and where it misleads, creative testing cadence, audience and exclusion strategy, incrementality and geo tests, landing page alignment, budget pacing, and defending spend to finance.

Produced by PPC.co, paid search and PPC management. Full details, services and further reading at https://ppc.co