← All posts

MCP: how I automated an entire Meta Ads operation without leaving the chat

3 min read
  • mcp
  • automation
  • ai
  • meta-ads
  • productivity

If you work with LLMs (Claude, ChatGPT, Gemini), you’ve probably been there: you’re in the middle of a productive conversation with the AI, you need to pull data from an external system, and then… you open another tab. Copy. Paste. Switch back. Lose context.

MCP (Model Context Protocol) solves exactly that. And when I say “solves,” I mean it completely changed the way I work.

What MCP actually is

No jargon: MCP is a way to connect APIs from any service directly to your LLM. Gmail, Google Calendar, Slack, databases, ad APIs — anything with an API can become a tool that the AI uses without you leaving the chat.

Instead of copying and pasting between tabs, you talk to the AI and it executes. Think of it as an assistant with access to all your systems.

What I built

I was running Meta Ads operations at scale — six-figure monthly budgets. The process of launching a new campaign involved:

  1. Finding the approved creatives from the team
  2. Defining headlines and primary texts (or creating variations)
  3. Configuring campaign parameters (audience, budget, placement)
  4. Following the naming convention (to keep the ads manager organized)
  5. Reviewing everything before publishing

Each new campaign took 30-40 minutes of repetitive work. It wasn’t hard — it was tedious. And tedious breeds mistakes.

So I built an MCP that did all of it.

How it worked

I’d open Claude and type /ad. From that point on, the entire pipeline was automated.

The AI already knew the path to fetch the creatives. Already knew which headlines and primary texts to use — and when to create variations. Knew the right settings for each creative type. Knew the naming conventions to follow.

Everything automated. Every step. From start to finish.

When it didn’t know something — for example, a creative decision that depended on context — it asked me. We always had a guardrail, a threshold to prevent errors. My last step was always the same: review and approve.

30-40 minutes became 5 minutes of review.

What it freed up

Time. But not idle time — productive time. Instead of configuring campaigns by hand, I could:

  • Analyze performance of running campaigns
  • Build new automations for other processes
  • Give more detailed feedback to the creative team
  • Think strategy instead of executing operations

This is the point many people miss about automation: it’s not about doing less. It’s about doing what matters.

Why MCP and not a traditional script?

I could have built a Python script that does the same thing. But MCP has one advantage that changes everything: natural language.

With a script, you need to anticipate every scenario, every exception, every edge case. With MCP + LLM, the AI adapts. If the creative is different from the standard, it asks. If information is missing, it searches. If something doesn’t make sense, it flags it.

It’s the difference between a robot that follows rules and an assistant that understands context.

How to get started

If you already use an LLM daily, MCP is the natural next step. Start simple:

  1. Identify a repetitive task that involves external systems
  2. Check if the service has an API
  3. Create an MCP server that connects that API to your LLM
  4. Test, adjust the guardrails, and use it

You don’t need to automate everything at once. One process. One workflow. If it works, you’ll never go back.