Why I Build My Own MCP Servers for Google Ads and LinkedIn
I got tired of exporting ad data to look at it. So I built MCP servers for Google Ads, LinkedIn Ads and Google Sheets, and now an AI watches the accounts with me. Here is why that is worth doing, and what it actually changed.
Three MCP servers now run against Google Ads, LinkedIn Ads and Sheets.
The Problem Was Never the Analysis
Every paid media person I know has the same weekly ritual. Open the platform, export a report, paste it somewhere, stare at it, decide what to change.
The analysis was never the hard part. The hard part was that it only happened when I sat down to do it. An account can burn three days of budget on a broken ad set between Tuesday and Friday, and nobody finds out until someone opens a dashboard.
That gap between "something changed" and "somebody noticed" is where most wasted ad spend lives.
What an MCP Server Actually Is
MCP stands for Model Context Protocol. In plain terms, it is a way to give an AI assistant a set of tools it can call on your behalf.
Instead of me exporting a Google Ads report and pasting it into a chat window, the assistant queries the account directly. It can pull campaign performance, search terms, change history, and conversion actions, and it can make changes when I tell it to.
The important part is that the tools are mine. I wrote them, I decide what they can reach, and I decide what they are allowed to change.
What I Built
Three of them now run in production for my own work.
Google Ads. Campaign and ad group performance, search term reports, change history, keyword management, budget and bid adjustments. It is deployed on Cloud Run and connected straight to my assistant.
LinkedIn Ads. Campaign performance, audience demographics, lead gen form results, creative performance. LinkedIn's reporting interface is the weakest of the major platforms, so this one saved the most time immediately.
Google Sheets. Less glamorous and possibly the most used. Half of client reporting still lives in a spreadsheet somewhere, and being able to read and write those directly closes the loop.
The Thing That Actually Changed
I expected to save time on reporting. That happened, and it is not the interesting part.
The real change is that the accounts are now watched continuously rather than in weekly bursts. I can ask a question at 9pm on a Sunday and get a real answer from live data, and I can set up checks that surface a problem when it appears instead of when I next look.
A search term that started eating budget on Wednesday gets caught on Wednesday. That one shift has been worth more than every hour of reporting time I saved.
Why Owning It Matters
There are off-the-shelf tools that connect AI to ad platforms. I looked at them. The reason I build my own comes down to the data.
When I run the server, the credentials are mine, the data path is mine, and client account data does not pass through a third party's infrastructure on its way to an answer. For anyone managing accounts they do not own, that distinction matters, and it gets harder to hand-wave every year.
There is a second reason. When the tool is mine, I can add the specific thing I need on a Tuesday afternoon instead of filing a feature request and waiting two quarters.
The Parts That Were Harder Than Expected
Authentication is most of the work. Google's OAuth flow, developer token approval, and the difference between a manager account and a client account will take longer than the actual reporting logic. Budget for that.
The API is not the interface. Things that take one click in Ads Manager can take three API calls and a resource name you have to look up. The mental model is different and it takes a while to stop fighting it.
Write access deserves fear. Reading data is safe. Changing budgets and statuses is not, and an assistant that can pause campaigns can pause the wrong campaign. I gate anything that spends money or changes state behind explicit confirmation, every time.
Should You Build Your Own?
If you manage ads at any real scale and you are comfortable reading API docs, yes. The leverage is genuine and the ceiling is high.
If you are managing one small account, probably not yet. The setup cost is real and the off-the-shelf connectors will get you most of the way.
The deciding question is not technical skill. It is whether you care who else touches the data on its way to the answer. If that matters to you, building it yourself is the only version that actually solves it.
If you want help working out whether this applies to your business, that is a conversation worth having.
More Notes
AI ToolsWhy I Run Some AI Locally and Some in the Cloud
Local models handle the boring half of my work and a frontier model handles anything genuinely hard. Here is where I draw that line and why AI does not actually give you more time.
AI ToolsWhich AI Model Actually Makes Sense for Running a Business
Claude, ChatGPT, Gemini and Meta Muse, judged on business work rather than benchmarks. My honest read after using all four, including the one I would recommend to most owners.
AI ToolsMeta Muse Review: OpenAI Finally Has a Real B2C Competitor
Meta has spent two years losing the AI race. Muse is the first thing they have shipped that beat every personal assistant I tried to build myself, and it took 90 seconds to set up.
AI ToolsThe Best AI Tools for Students: 6 That Are Actually Useful
Most lists of AI tools for students are affiliate spam for products nobody uses twice. These six are the ones I would actually recommend, split by what they are genuinely good at and where they fail.