Claude with the Meta MCP vs a bulk ad uploader: which one launches 60 ads a week?

Yes, an AI assistant connected to Meta's MCP can create ads. For drafting a campaign that is a real step forward. For launching the same 60 ads every week, exactly, it is the wrong tool, and the reasons are mechanical, not philosophical.

Adlio team4 min read

The short answer

An AI assistant with the Meta MCP is excellent at the parts of advertising that need judgment: which audiences, which angles, what the last month of results means, five headline variants for a new offer. A bulk uploader is built for the part that needs none: taking 60 finished creatives and turning them into 60 correctly configured ads in Ads Manager, this week and every week after.

The split is not "AI vs software". It is variable work vs repeated work. Language models are strong at the first and unreliable at the second, and launching ads in volume is the second.

What the assistant does well

  • Drafting. "Give me eight hooks for this product for a cold audience" is a five-minute conversation that used to be an hour.
  • Reading data. Pull the last 30 days through the MCP and ask what changed; you get a decent first read without opening a spreadsheet.
  • One-off structure. Creating one campaign with three ad sets and specific targeting by describing it in plain language works, and it is faster than clicking through the interface for a structure you will build once.
  • Explaining Meta. Ask why an ad set is stuck in learning or what an error code means and you get a useful answer.

None of that is what a bulk uploader does, so nothing here is a reason not to use the assistant.

Where it breaks on the weekly batch

Five mechanics, each one enough on its own.

1. One ad per turn

The MCP exposes API calls; the assistant issues them one at a time, in a conversation. Sixty ads is sixty creative objects and sixty ad objects, each waiting for the previous one, each a chance for the model to drift. An uploader issues the same calls from a loop with the same parameters, in parallel, with retry logic. The difference is not speed alone; it is that the sixtieth ad is configured exactly like the first.

2. Enhancements are a per-creative opt-out, and the opt-out is a payload

Advantage+ creative enhancements are stored on each creative and are on by default. Turning them off through the API means sending a degrees_of_freedom_spec with the exact verified keys, each set to OPT_OUT, on every creative you create. Send a key Meta does not accept and the whole creative fails. A model has to reproduce that block perfectly, sixty times, from memory of an earlier instruction. A tool ships it as code, and only ships keys verified against the live API. The full mechanics are in how to turn off Advantage+ creative enhancements.

3. Media is not a chat message

Your creatives are files: 40 images and 20 videos, some of them 1 GB. They have to be uploaded to the ad account, videos processed, IDs collected, then referenced. That is a file-transfer job with resumable uploads and status tracking, not something you paste into a conversation. See bulk uploading video ads to Meta for what that path involves.

4. Naming and UTMs need to be identical, not similar

A naming convention works when every ad follows it character for character. An assistant asked to "name them like last week" will produce names that are close. Close is what breaks a reporting breakdown. A pattern with tokens produces the same string every time; the naming convention guide covers why that matters.

5. Safety has to be structural

An assistant with write access to your ad account can create an active ad if the instruction is ambiguous. That is not a criticism of any particular model; it is what "general purpose" means. An uploader that cannot create an active ad, by design, removes the failure mode instead of asking you to prompt carefully. Adlio creates every ad paused; there is no setting to change that. The pre-launch QA checklist is built around it.

Cost, briefly

Every ad created through an assistant consumes tokens for the reasoning and the tool calls around it, and the cost scales with the number of ads and with how often the model re-reads context. A flat-priced uploader costs the same for 15 ads or 500. We are not going to quote token prices here because they change monthly; the shape of the curve is the point.

The workflow that actually works

  1. Think with the assistant. Angles, hooks, audiences, copy variants. Export the copy you approve.
  2. Produce the files. Name them by your convention at export time (SUMMER_vid_02_9x16.mp4).
  3. Ship with the uploader. Drop the folder, pick the campaign and ad set, launch paused, review, set live.
  4. Read the results with the assistant a week later and go back to step 1.

That is also why we treat "draft in Claude, launch through Adlio" as a workflow to support rather than a competitor to fight. The two tools want different halves of the job.

Try the shipping half

Adlio is free for your first 15 ads, with no card and no trial clock. Connect Facebook, set your account defaults once (Page, pixel, naming, UTMs, enhancements off), drop a folder. See how bulk upload works and pricing or the FAQ.

Frequently asked questions

Can Claude or ChatGPT create Meta ads directly?+

Yes, through Meta's MCP server (or any tool that wraps the Marketing API), an AI assistant can create campaigns, ad sets, creatives and ads in an ad account you connect. It does so one API call at a time, driven by the conversation.

Will an AI assistant turn Advantage+ enhancements off reliably?+

Only if it emits the exact creative_features_spec keys with enroll_status OPT_OUT on every single creative, every time, and never includes a key Meta rejects. A language model can do that on Tuesday and forget on Thursday. A tool does it because the opt-out is code, not a prompt.

Is Adlio itself available as an MCP server?+

Not today. We are looking at exposing the launch job to AI agents, because "draft in Claude, launch through Adlio" is a workflow we like. Until then Adlio is a web app: connect Facebook, drop the folder, review the paused batch.

Which should a solo media buyer pick?+

Both, for different jobs. Use the AI assistant to think: audiences, angles, copy variants, reading your reports. Use an uploader to ship: the batch of files that has to land in Ads Manager correctly named, with enhancements off and paused. The mistake is asking the assistant to also be the shipping department.

Adlio team

Built by media buyers who launch Meta ads in volume. Questions or corrections: support@adlio.ai.

Topic cluster

Tools and alternatives

Honest comparisons of the ways to bulk launch Meta ads, free and paid.

Start here

The best bulk ad launchers for Meta ads

Related guides