Anyone else's agency or in-house team moving to MCP-based PPC automation?
Running a small Amazon ads agency, and the last 3 months have felt like the workflow is quietly shifting under us.
Quick context for anyone who hasn't seen the term - MCP (Model Context Protocol) is basically a standard way for AI models to talk to external tools.
Instead of copy-pasting a search term report into ChatGPT and asking it what to negative out, the model itself can pull the report, cross-reference it with conversion data, and propose (or execute) the changes.
What I'm seeing on our side:
- Search term harvesting and negative targeting that used to be a weekly Friday job is now running closer to daily, because the marginal cost of doing it is near zero once the MCP connection is set up.
- Bid adjustments at the keyword level - the kind we'd batch into Bulk Sheets once a week , are getting done in smaller, more frequent passes. Not because we want to micromanage, but because the model can actually keep up now.
- Reporting is the one that surprised me. Pulling weekly numbers across 10+ accounts used to eat half a day. It's now mostly a review step, not a build step.
The thing that changed for me is Gemini 3.5 Flash dropping today.
Speed and cost finally got to the point where running an agent continuously in the background isn't a hypothetical anymore. It's just a Tuesday.
A few things I'm genuinely unsure about and would love other operators' takes on:
- How are you handling the trust layer? We still have a human approve anything above a certain bid delta or budget change. Curious where others draw that line.
- Are clients noticing? Mine mostly haven't, which is either a good sign (it's working) or a bad sign (we're not communicating the change well).
- For the larger agencies here - at what account count does this stop being a nice-to-have and start being the only way to keep margins?