Run your ads from Claude Code.

What this actually is

Claude Code is Anthropic's agent that runs in a terminal. It supports remote MCP servers, which means it can call tools hosted somewhere else. Muze is one: 149 paid media tools covering Meta Ads at 37, Google Ads at 41, Amazon Ads at 48, Shopify at 15, and 8 cross-platform tools including get_blended_performance.

Once connected, the ad account is addressable from the command line. You can ask for last week's wasted spend the same way you would ask for a failing test, and you can ask for it again next week and get the same shape of answer. That repeatability is the actual feature. A chat answer varies with how you phrased the question; a job you run the same way every Monday does not.

Who this is actually for

Agencies and in-house teams running the same checks across many accounts. The work of managing twelve client accounts is not twelve interesting strategy conversations, it is the same five reads repeated twelve times, and that is exactly the shape of work a terminal agent is good at.

It also suits anyone who wants the output somewhere other than a chat window. Claude Code can put the result into a file or a commit, so the account audit stops being something you read once and becomes something you keep. The safety model does not loosen because you moved to a terminal: writes still preview, campaigns are still created paused.

How connecting works

You need a Muze account with at least one connected ad or commerce account, and a Muze API key if your setup authenticates with one. Platform access is always OAuth, done once inside Muze.

1. Create a Muze account and authorize your accounts. Sign up for Muze and connect Meta Ads, Google Ads, Amazon Ads or Shopify over OAuth. This is the only place platform access is granted, and you can revoke it at any time.
2. Create a Muze API key. Generate a key on your Muze API Access page. Keys begin with mz_live_ and are shown once, so store it somewhere sensible. Rotate it if it is ever exposed.
3. Add Muze as a remote HTTP MCP server. Point Claude Code at https://backend.muzecmo.com/mcp as a remote HTTP MCP server, with your Muze API key in the X-API-Key header. The connection guide has the exact configuration entry.
4. Reload and verify. Reload the MCP server list, then ask Claude Code to list your connected ad accounts. If get_my_accounts returns them, you are connected. If the tools are missing, reconnect so the client refreshes its cached tool list.

This page deliberately does not print a command line for the install step. Client syntax changes between versions, and a stale command is worse than none.

The connection guide carries current per-client configuration detail.

What you can ask once it is connected

Terminal work rewards jobs with a defined output. Say what you want ranked, over what window, and where the answer should end up.

• Audit all connected accounts for the last 7 days and write the findings to a markdown file. get_my_accounts, then get_meta_performance, get_google_performance and get_amazon_performance.
• For every client account, list campaigns pacing more than 20 percent over budget. list_campaigns_detailed and list_amazon_campaigns across each authorized account.
• Pull Amazon search terms with spend and zero orders, and draft the negative keyword list. list_amazon_search_terms, then create_amazon_negative_keywords behind a confirmation.
• Compare blended MER week over week and tell me which platform moved it. get_blended_performance twice, with get_shopify_revenue for the denominator.
• Check how many tool calls I have left this month before I run the full sweep. get_my_usage, which is worth calling first when you are about to loop over accounts.

Those are real tool names, not illustrations.

Nothing spends until you say so

An agent that only reports is a nicer spreadsheet, so Muze can act. The gate in front of it is what makes that safe.

• Every write previews first. You see exactly what will change, and it waits for your confirmation.
• New campaigns are created paused. Creating one cannot start spending.
• Access is scoped to the accounts you authorize over OAuth, and you can revoke it.
• Muze never takes a percentage of your spend. You pay for software, and your budget stays yours.

Running in a terminal does not make writes autonomous. Every write still previews and still waits for an explicit confirmation, and campaigns are still created paused. Automation here means repeatable, not unattended.

Where Claude Code is the wrong client

There are no charts. If the deliverable is something a client looks at, you will be rebuilding it elsewhere. A terminal is a good place to find the problem and a poor place to present it.

It is also a developer tool with a developer's setup cost, including an API key in a config file that must not end up in a repository. If the person who owns the account does not live in a terminal, put Muze in ChatGPT or Claude instead. And Claude Code does not schedule itself: repeatability comes from you running the same prompt.

One connection, four platforms

Meta Ads, Google Ads, Amazon Ads and Shopify all arrive through the same server, which matters more in a terminal than anywhere else. A sweep across every client account and every platform is one loop, not four integrations with four auth models and four sets of rate limits.

Platform detail: Meta Ads, Google Ads, Amazon Ads, Shopify. The Muze MCP overview has the tool inventory, pricing has the plans, and every supported client is listed in one place.

Frequently asked questions

How do I add Muze to Claude Code?

Add it as a remote HTTP MCP server pointed at https://backend.muzecmo.com/mcp, authenticated with your Muze API key in the X-API-Key header, then reload the MCP server list. The connection guide carries the exact configuration entry.

Can it run unattended, on a schedule?

Reads can be repeated as often as your plan's tool call allowance permits. Writes cannot run unattended by design: every one previews and waits for confirmation. If a workflow needs a budget changed at 3am with nobody watching, Muze is the wrong tool.

Does the terminal client get fewer tools?

No. All 149 tools are the same across every client: 8 cross-platform, 37 Meta, 41 Google, 48 Amazon and 15 Shopify. What differs is the setup path and the presentation, not the capability.

Where do I get an API key?

From the API Access page in the Muze app. Keys start with mz_live_ and are displayed once. If a key leaks, rotate it there. Rotating a key does not disturb the OAuth connections to your ad platforms.

Is this good for managing multiple client accounts?

It is the best client for it. One connection covers every account you authorize, and the same audit prompt runs across all of them without opening a dashboard.

What does it cost?

The free plan includes 25 read-only tool calls a month. Write actions require a paid plan. Muze charges a flat subscription and never a percentage of ad spend, so a busy month costs you the same as a quiet one.

Muze in other AI clients