Your media buyer is going to hate you having this level of insight into your campaigns. You will have a system that is smarter than them and will save you thousands in wasted ad spend.

Over the last couple of weeks we've been using Claude to directly interact with our Meta ads and help improve performance across multiple client accounts. We connected Claude to Meta using something called an MCP server. In short, it means we can ask Claude questions in natural language, it pulls relevant data from Meta, and then suggests ways to improve performance for our campaigns.

It can analyze data, suggest changes, and even make the changes for you.

Here's what Claude discovered across different accounts we manage:

What we discovered

We started by asking Claude to audit several campaigns across different client accounts. Within minutes, it had identified critical issues that we'd been missing for weeks.

Here's what Claude found:

Real insights Claude found across our accounts

Account 1 - E-commerce Brand: Claude immediately flagged: "Critical Issue #3: Severe Ad Fatigue in Retargeting"

Account 2 - DTC Daily Newsletter: Claude identified the core problem with our newsletter ads: "You're building a premium B2B audience, not just collecting emails"

Account 3 - Ecommerce Brand: Claude recommended immediate budget reallocation:

These aren't theoretical suggestions - this is Claude analyzing real campaign data and finding money-wasting patterns we missed.

What this unlocks for you

You can talk to Claude like it's your media buyer & analyst in natural language:

  • "Show me all ad accounts in Meta"

  • "Compare ROAS week-over-week by campaign"

  • "Pull top 5 creatives with thumbnails and CTR"

  • "Pause anything spending >$100 with ROAS under 1.2"

  • "What age groups are wasting my budget?"

  • "Create a new campaign targeting B2B decision makers"

  • "Which campaigns have dangerous frequency levels?"

  • "Reallocate budget from underperformers to winners"

No spreadsheets. No screen-sharing. No screenshots.

Claude sees the data and can make changes based on your instructions.

The Setup

The setup takes only 15 minutes. No code or engineer required.

  • Step-by-step installation guide (built for non-technical folks)

  • Exact files to paste

  • Over 20 Prompt for audits, optimizations, and actions

You'll save 5–6 hours/week, at minimum.

Step-by-step workflow for Claude + Meta Ads MCP

If you want Claude to act like your personal media analyst, this is how you set it up. No complex API setups or dev work needed. Just local tools that plug into your Meta Ads data.

What this setup lets you do:

With Claude connected to Meta Ads via the MCP (Media Control Panel), you can:

  • Pull campaign and ad set performance data on demand

  • View thumbnails, CTRs, ROAS, spend, and more

  • Ask questions like: “Which campaigns had a ROAS drop last week?”

  • Take action: “Pause all ad sets spending >$100 with <1 ROAS”

Meta Ads MCP Demo.mp4:

These instructions are for Mac users. If you’re on Windows or Linux, you can copy the entire instructions and just ask ChatGPT: “Convert this setup guide from Mac to Windows/Linux.” It’ll rewrite the steps for your system in seconds.

And because of how it’s set up:

  • Your data is pulled securely via Pipeboard

  • MCP runs locally on your machine

  • You stay in full control of access

What you’ll need

  • A MacBook or an iMac

  • Claude Desktop (from Anthropic)

  • Terminal (built into every Mac: press Cmd + Space and search “Terminal”)

  • Pipeboard API token (this gives secure access to your Meta Ads account)

  • Meta Ads MCP tool (open-source connector that talks to Claude)

Step-by-step setup

Step 1: Install Homebrew (if you haven’t already)

Homebrew is a free tool that helps you install other tools safely. It's like an app store for your terminal.

In Terminal (Press Cmd + Space to open Terminal on MacOS), paste this:

  • You might be asked to enter your Mac password.

  • Type it (it won’t show that you’re typing something, or have typed something, and that’s normal) and press Enter.

  • Let it run for 5–10 minutes.

  • You’ll know it’s installed when you see messages like “Installation successful” or a prompt returning to your terminal (no errors). If nothing happens after 10 minutes or it gets stuck, close Terminal, reopen it, and repeat the command.

Then, paste these lines to finish the setup.

If your terminal says zsh (check the top of your terminal window), paste this:

echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile

eval "$(/opt/homebrew/bin/brew shellenv)"

If your terminal says bash, use this instead:

echo >> ~/.bash_profile

echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.bash_profile

eval "$(/opt/homebrew/bin/brew shellenv)"

Example: For my terminal, it says Zsh, as you can see in the screenshot

Step 2: Install uv (universal virtual environment tool)

Now that Homebrew is working, install uv. Think of uv like the engine that powers your local tool (MCP). It helps manage environments and packages cleanly, so the Claude connector runs smoothly without breaking other parts of your system.

This tool is trusted and open-source. You're not installing anything risky, just a clean, local helper that Claude talks to.

Paste this into Terminal:

brew install uv

Once done, test that it worked:

If the command works, you'll see a version number like 4.5.7 or similar. That confirms uv is installed correctly.

If you see command not found or an error:

  • Close Terminal and reopen it, then try again.

  • Make sure you completed the Homebrew setup steps properly for your terminal shell (bash or zsh).

  • If it's still not working, paste the Homebrew setup commands again, then retry the install.

Enter what’s below, and press Enter. You won’t see a big “success” banner, but just the version number. That’s how you know it worked.

uv --version

You should see a version number like 4.5.7 or higher.

Step 3: Install the Meta Ads MCP tool

This is the package that connects your Claude to Meta Ads securely.

In Terminal, run:

pip3 install meta-ads-mcp

This might take 1–2 minutes.

Step 4: Get your Pipeboard API token

  • Create a free account

  • Click "Create Token"

  • Copy the token string (starts with pk_)

It will look something like this when your Token is generated.

This token is how MCP gets secure access to your ad account.

Step 5: Launch the MCP server locally

Paste this into Terminal (replace with your actual token):

export PIPEBOARD_API_TOKEN=pk_your_actual_token_here

uvx meta-ads-mcp

If it’s working, you’ll see something like:

Uvicorn running on http://127.0.0.1:8000

This means the server is live on your machine.

Step 6: Connect Claude to the MCP server

  1. Open Finder

  2. Press Cmd + Shift + G

  3. Paste this path:

~/Library/Application Support/Claude/

  1. Inside this folder, you’ll see a file called config.json.

  1. Right-click it and select Duplicate.

Rename the file to:

claude_desktop_config.json

6. Paste this JSON code inside claude_desktop_config.json (replace with your real token):

JSON

Copy

{

"mcpServers": {

"meta-ads": {

"command": "uvx",

"args": ["meta-ads-mcp"],

"env": {

"PIPEBOARD_API_TOKEN": "pk_your_actual_token_here"

}

}

}

}

Save and close.

Step 7: Restart Claude Desktop

  • Quit Claude completely

  • Open it again

Now ask:

"Show me all my ad accounts."

Claude will now respond with your actual Meta Ads data.

20 prompts you can run instantly inside Claude once it starts running:

  1. "Show me all my Meta ad accounts"

  2. "List all campaigns that ran last month sorted by ROAS"

  3. "Which ad sets had CTR below 0.8% in the past 7 days?"

  4. "Pause ad sets spending more than $100 with ROAS under 1"

  5. "Pull thumbnail previews for top 5 creatives by spend"

  6. "Summarize daily performance for [campaign name]"

  7. "What was the total spend across all active campaigns last week?"

  8. "List all campaigns with negative week-over-week ROAS change"

  9. "Which ad sets are burning budget with no conversions?"

  10. "Sort all active ads by CPA from low to high"

  11. "Generate a daily Slack report summarizing campaign health"

  12. "Export campaign performance from the past 30 days to CSV"

  13. "Create a new naming convention report for all active ads"

  14. "Flag any ad sets that haven’t spent in the last 3 days"

  15. "Which audience segments are performing best by ROAS?"

  16. "Show breakdown of spend by device placement"

  17. "Rank creatives by thumb stop ratio"

  18. "Highlight anomalies in spend spikes week-over-week"

  19. "Summarize campaign results for investor update deck"

  20. "List all new campaigns launched in the last 14 days"

Keep Reading

No posts found