Managing Google Ads campaigns with ChatGPT: eight jobs through a remote MCP server

ChatGPT cannot see your account on its own; it needs a bridge. MCP, the Model Context Protocol, is that bridge: the server Tecrube hosts offers ChatGPT tools such as “pull report”, “add negative” and “change budget”, and ChatGPT calls them according to what you ask in chat.

You do not run a server. In ChatGPT you enable developer mode under Settings → Connectors, add the address copied from the panel as a custom connector and link your Google account with OAuth. No password is shared and no API key is needed.

Step one: add the connector to ChatGPT; this is done in the web interface on paid plans only. Step two: type “connect my Google Ads account” in chat and grant access in Google's window; the grant is tied to your Tecrube account and done once. Step three: start using it. The same address also works in Claude, Claude Code and Cursor.

Once setup is done, ask “list my Google Ads accounts” to confirm you connected the right one; if several appear, state at the start of the chat which one you want to work on.

1. Performance report: spend, conversions and CPA by campaign, ad group and keyword, with a period comparison. 2. Wasted search terms: terms that click without converting, sorted by spend. 3. Keyword quality: keywords with Quality Score below 5, with their components. 4. New keyword ideas: with volume and bid range, grouped by intent.

5. Budget optimisation: shift suggestions from capped campaigns to ones that do not spend. 6. Change history: who changed what in the past week. 7. A new campaign: ad groups, keywords, RSAs and extensions in one preview. 8. Account audit: conversion tracking, structure, ad strength and budget findings in priority order. Most of the first six are reads and spend no credits.

When you say “raise the Search campaign budget to $60 and add ‘second hand’ as a negative”, Tecrube produces two separate previews: current value, new value, which campaign. What you approve is applied and written to the audit log; what you do not approve stays as it was.

Single-step budget increases above fifty percent are rejected. Because the approval gate is on the Tecrube server rather than in ChatGPT, client settings cannot bypass it.

One message on Monday morning: “summarise last week, list the wasted search terms, show the campaigns limited by budget”. Three answers arrive in one reply. Then two approvals: adding negatives and shifting budget. Fifteen minutes in total, where this used to take an afternoon.

A full audit once a month and fresh keyword research once a quarter round out the routine.

The free account gives 75 credits that never expire; no card is needed. Pro at $49 a month gives a single business one account per platform and 5,000 credits; Agency at $99 a month gives unlimited credits and every client under an MCC; Enterprise starts at $299. Looking at the panel spends no credits and the connection is removed in one click.

A routine like the weekly example above means a few dozen writes a month; for a single business, Pro's 5,000 monthly credits cover that with plenty to spare.

The quality of the assistant's answer depends on how clear your question is. Three elements always make a difference: date range, criterion and output format. Instead of “how are my campaigns?”, saying “show campaigns with CPA above $25 over the last 14 days in a table sorted by spend” gives a usable answer in one go.

Give the standing context once: what your business sells, your target cost per conversion, which cities you serve. If you put this into ChatGPT's project or custom instructions features, you do not need to repeat it in every chat and the suggestions are shaped around your goal.

Finally, in write requests say what you do not want as well: a boundary such as “do not change the budget, only suggest negatives” keeps the changes that reach the preview exactly within the scope you intended.

Put together, these three habits turn Google Ads management with ChatGPT from a chat that gives generic advice into a working routine that knows your account, goals and limits and presents every change as a preview. Adjusting your instructions a few times in the first weeks is normal; recurring misunderstandings in the assistant's answers show where context is missing.