What it can do to your ad accounts, and what it cannot.

The eight

Why these eight and not a list of forty

Feature lists are easy to write and impossible to verify, which is why every AI marketing site has one and none of them mean anything. These pages work differently. Each capability on them maps to named tools in the MCP inventory, and where the Muze app can do something the MCP cannot, the page says so rather than letting the sentence blur.

The clearest example is creative. Muze generates images and video inside the app, but those generators are not MCP tools, so the creative page describes what the protocol actually exposes: assembling creatives, swapping the image or video on a live ad, and rewriting copy. Similarly, autopilot runs on Google and Amazon, and not on Meta. Neither of those is the flattering version. Both are the true one.

Read, then write

Most of these features are read-only, and that is a deliberate ordering rather than a limitation. Cross-platform analysis, blended reporting and the detection half of wasted spend change nothing, so you can run all of them on the free tier before deciding whether you trust the product with a write.

When you do write, the model is consistent across all four platforms. Every change previews and waits for explicit confirmation. Campaigns, ad sets and ads are created paused, and the paused status is forced by Muze’s API rather than chosen by the assistant. Thirteen tools with the most consequences, the pausing, resuming, archiving and disconnecting ones, require explicit confirmation before they will execute. Read how approval controls work if that is the part you care about, which it should be.

Where to start

• Spending on more than one platform: cross-platform analysis, because it is read-only and it is the question a single dashboard cannot answer.
• Suspicious that budget is leaking: wasted spend detection, then the guide on how to find wasted ad spend.
• Running a Shopify store: blended reporting, which puts all your ad spend against real order revenue.
• Evaluating whether this is safe at all: approval controls, and then how the connection works.

Frequently asked questions

What can Muze actually do to my ad accounts?

Read everything, and write with your approval. Across Meta Ads, Google Ads, Amazon Ads and Shopify, Muze exposes 149 tools: 71 read, 65 write and 13 classed as destructive. Reads cover performance, campaigns, ads, search terms, competitor ads and Shopify orders. Writes cover campaign creation, budgets, bids, pausing, negatives and creative swaps.

Which platforms does the Muze MCP support?

Four: Meta Ads with 37 tools, Google Ads with 41, Amazon Ads with 48 and Shopify with 15, plus 8 core cross-platform tools. It does not support TikTok, Bing, LinkedIn or X. The wider Muze platform covers more channels than the MCP does.

Is it safe to connect an AI to a live ad account?

That depends entirely on the write model, which is why it is a feature page of its own here. Every Muze write previews first and waits for explicit confirmation, new campaigns are created paused with the status forced server side, and the 13 highest-consequence tools are classed as destructive and require your explicit confirmation before they will run.

Do I need a developer account or API access of my own?

No. Connection is OAuth with your normal platform logins. There are no API tokens to mint and no platform passwords to hand over, and access is scoped to the accounts you authorize and revocable at any time.

What does it cost?

The free tier is 25 read-only tool calls per month across all connected platforms. Write actions require a paid plan. Muze never takes a percentage of ad spend, so your budget stays in your own ad accounts.

Does everything work in ChatGPT and Claude?

Yes, and in any other MCP client. The tools are the same regardless of which assistant is calling them, because the capability lives in the connection rather than in the chat app.