No Ads Experience Needed: One AI Agent for Creative + Campaigns

One-liner: A Meta ad traditionally takes two people — a designer for the creative, a media buyer for the campaign. This playbook collapses both into one AI agent + two MCPs. And you don’t need to be technical or know how to run ads: you just talk and approve; the agent does the work. The whole thing can run at $0 — the only moment real money moves is when you say “go.”

The 4 things you (the boss) actually do

This sounds like an engineering task. It isn’t. From start to finish you only do four things:

Connect AutoWhisper to your AI agent (via skill or MCP — pick one, set up once).
Connect Meta’s official Ads MCP to your AI agent (set up once).
Tell the agent what you want, in plain language — this playbook writes out every sentence you need to say.
The agent does it; you review — it builds the whole campaign (paused, $0), sends you a preview, and nothing goes live until you’re happy with it.

You never touch a “command line” and you never click through the maze of menus in Meta Ads Manager. You say things and you make calls. The agent handles the rest.

What to have ready first

The first time you set this up, get these together (most are one-time):

A Facebook account + a Facebook Page. Every Meta ad runs “as” a Page, so without one you can’t advertise. If you don’t have a Page yet, create one on Facebook (“Create a Page” — a brand name + a photo is enough to start).
A Meta ad account with a payment method attached. Set this up in Meta Ads Manager (business.facebook.com). Note: building the campaign objects costs nothing, but the account has to be able to spend, or the final “go” step will stall.
An AutoWhisper account with your product set up and some content already generated (feature posters / lookbooks / videos). New accounts get free credits — you don’t have to pay to test this end to end.
- Where the token is: log into autowhisper.xyz → Settings → “Connect your agent.” There’s an API token on that page — copy it. That token is the key you’ll use to connect AutoWhisper; save it somewhere.
An AI agent that supports MCP: Claude Desktop, Cursor, Cline, Claude Code — whichever you’re comfortable with.

Setup step 1: Connect AutoWhisper to your agent

Pick one. You do this once.

Option A: MCP (recommended). Add this to your MCP client config (replace YOUR_TOKEN with the token you copied):

{
  "mcpServers": {
    "autowhisper": {
      "command": "npx",
      "args": ["-y", "autowhisper-mcp"],
      "env": { "AUTOWHISPER_API_TOKEN": "YOUR_TOKEN" }
    }
  }
}

⚠️ It must be autowhisper-mcp 0.1.1 or newer — older versions won’t hand over the creative’s image URLs, and you’ll get stuck at the “get the links” step. npx -y autowhisper-mcp pulls the latest automatically, so you usually don’t have to think about it.

Option B: Skill (good for Claude Code / clawdbot-style environments):

npx skills add xnjiang/autowhisper-skill
mkdir -p ~/.config/autowhisper
echo '{"api_token":"YOUR_TOKEN"}' > ~/.config/autowhisper/credentials.json

After either option, restart your agent client so it picks up the new connection.

Setup step 2: Connect Meta’s official Ads MCP

This is the one slightly technical part of the whole flow — but it’s also once. Two paths, pick one:

Easiest (recommended if you’re not technical): use the built-in connector. Many agent clients ship a “Facebook / Meta Ads” connector. Find it in settings, click “Connect / Authorize,” log in with your Facebook account, and you’re done — nothing to memorize.

Command line (Claude Code, etc.):

claude mcp add --transport http --client-id <YOUR_META_APP_ID> meta-ads https://mcp.facebook.com/ads

This one needs you to have your own Meta app at developers.facebook.com and paste in its “App ID.” If you’re not technical, don’t fight this — use the built-in connector above, or have a technical friend set it up once (after that it just works).

Once connected, you don’t need to remember any tool names. Meta’s whole set of ad-building tools (create campaign, create ad set, render preview, and so on) is something the agent knows how to call — you just describe what you want in plain language.

Setup step 3: Tell the agent what you want (say these lines)

For each step below I’ve written out the exact thing to say to your agent. Say it as-is, or reword it — the agent will call Meta’s tools behind the scenes and build everything for you.

① Have it generate the creative

You say: “Generate a few feature posters and lookbook images for my product AutoWhisper.”

⚠️ Gotcha 1 — generation takes a few minutes (it’s async). Creative isn’t instant; it takes a few minutes to render. The agent will very likely reply “generating in the background” the first time — that’s normal, not a failure. Don’t nag it or retry; wait a few minutes before the next step.

② Have it pull the image links for the creative

You say: “Are they done? Give me the image links for AutoWhisper’s feature posters.”

⚠️ Gotcha 2 — name the product (“AutoWhisper”). One AutoWhisper account often has several workspaces (say, a “Default” full of demo products, and another one that’s your real product). If you don’t name the product, it defaults to the first workspace and probably hands back a pile of unrelated creative. Name the product and it’ll locate the right one across all your workspaces.

⚠️ Gotcha 3 — asset type decides whether it can go straight into an ad:

Feature posters and lookbooks = static images you can use directly. The agent gets the link and pipes it straight into Meta.

Videos (voiceover UGC, story, product demo) = this interface can’t upload video. The fix for video is: let the agent build everything else, then upload the video yourself in Meta Ads Manager (its UI supports video in a few clicks). Don’t make the agent wrestle with it.

Trust posts = the cover is often a stock/placeholder photo, not your real product, so don’t use it as an ad image.

You can just tell the agent: “Only feature posters and lookbooks — no videos or trust posts.”

③ Have it build the ads (one sentence, all your requirements)

You say (one plain-language paragraph): “Using those feature posters and lookbooks, build a campaign in my Meta ad account that drives traffic to my site autowhisper.xyz. Target the US, Canada, the UK, Australia, and New Zealand. Keep the audience broad with Advantage+ on, don’t add manual interests. Put a few different images and angles in one ad set to test against each other. Keep everything PAUSED — do not launch. Name the campaign, ad set, and ads properly. When it’s built, send me the previews.”

Behind the scenes the agent builds the campaign, ad set, and each ad, naming them cleanly (something like AutoWhisper_Traffic_US_2026-07 / US_Broad_Advantage-Audience / Hook_StopChasingAlgorithms_FP32). You don’t have to remember these — just knowing they’re tidy and easy to review later is enough.

✅ Less to worry about: Meta now defaults to Advantage+, so most settings are already right — leave them alone. You used to hand-pick the audience, the placements, a pile of targeting. Now Meta optimizes the audience (Advantage+ Audience) and placements (Advantage+ Placements) for you by default — you and the agent don’t need to touch them; leaving the defaults is the optimal setup. The more you fiddle with these, the worse it gets — don’t get clever and narrow them. The one exception is the single switch in Gotcha 4 below: “turn off only that one, leave everything else.”

A couple of “expert defaults” are baked into that sentence — here’s why:

Why those 5 English-speaking countries? The same English creative works across all of them — this isn’t “spraying wide.” And Canada / UK / Australia usually have cheaper clicks than the US, which feeds the algorithm more data to “learn” who to show it to, faster.
Why “no manual interests”? Guessing geos and guessing interests are the same mistake — you think you’re being precise, but you usually box out the real buyers. Advantage+ Audience is literally built to find the people for you; adding manual interests is a step backward. Go broad, let the creative filter people, then narrow once you have data.
⚠️ Gotcha 4 — the one switch you actively turn off: “Don’t turn on Advantage+ Creative (auto-edits) — I want to test each image cleanly.” Don’t confuse this with the above: keep the Advantage+ audience and placements on — but “Advantage+ Creative” (the thing that auto-crops / adds filters / rewrites copy / swaps music) should be off during testing. With it on, you can’t tell whether a good result came from “the image itself” or “Meta’s edit,” and your A/B test turns to mush. Once you know which creative is the winner, then consider turning it on.

④ Review the previews (you make the call)

The agent gives you preview links one by one; open them to see how each ad looks in a real placement (say, the US feed).

Don’t like one? Just say: “Swap ad #2 for a lookbook.” “Change this copy to XXX.” “The headline’s too long, shorten it.” The agent redoes it and shows you again.
⚠️ Gotcha 5 — you haven’t spent a cent at this point. Building ads, editing creative, viewing previews — all of it is paused, all of it is $0. The only thing that spends money is when you say “launch.” Don’t trip that switch during testing.

Setup step 4: The calls you make before launching

Once you’re happy with the previews, a few things are yours to decide, not the agent’s:

Set the budget. You say: “Set the budget to $XX per day.” ⚠️ The key: the budget has to be enough for the ad to “learn” — an ad set needs roughly a few dozen conversions per week to stabilize. A budget that’s too low (a few dollars a day) basically means it never learns — no creative can save that. This isn’t “the ads missed,” it’s that there wasn’t enough money for the algorithm to learn at all. Better to run fewer days with a real daily budget than to spread it too thin.
(Recommended) Add tracking. You say: “Add UTM tracking to these ads with dynamic parameters that include the ad name.” Then AutoWhisper’s own attribution can tell you “which ad, which region drove the signup,” instead of you squinting at Meta’s dashboard.
DSA declaration (when targeting the UK / EU). Because the targeting includes the UK, Meta may pop up a box asking for the DSA “beneficiary / payer” when you hit publish — put in your business/brand name, don’t skip it.
After launch, it goes to review first. After you say “launch,” Meta reviews it (minutes to hours) before it actually runs. It occasionally gets bounced over the AI disclosure or a specific dollar amount in the copy — have the agent tweak and resubmit; it doesn’t mean the ad is bad.

Judgment discipline: don’t fool yourself

Once the ads are live, the easiest place to go wrong is reading the numbers. A few hard rules:

Cheaper clicks ≠ better results. After you add cheaper regions, your average cost-per-click usually drops — but half of that is because Meta went off to harvest volume in the cheap markets. It optimizes for “the cheapest click,” and can quietly pour budget into the cheapest country while starving the market you actually want. Judge on cost per signup, not cost per click. Clicks from low-intent cheap markets tend to convert worse.
Don’t narrow early on a hunch. Go broad, let the creative filter people, and once Meta gives you the region/age/gender breakdown, narrow by data, not by gut. The “precise audience” you imagine is often the wrong one.
Impressions and clicks aren’t progress. The number to watch is always how much you spent per signup / per real customer.
For most DTC categories, video usually converts best (native, doesn’t look like an ad). But as in Gotcha 3, video has to be uploaded to Meta by hand. Don’t default to static just because it’s easier to pipe in — the two extra minutes to upload a video can be the highest-return two minutes you spend.

Who this replaces

Back to the opening: a Meta ad used to take two people. This flow reassigns the work:

The designer’s job: creative is generated by AutoWhisper’s AI CMO — feature posters, lookbooks, videos, in 10 languages, and it looks at what competitors are running while it does.
The media buyer’s job: ad structure, naming, audience, previews — done by Meta’s official Ads MCP + the agent in one pass.

What’s left for you is judgment: how much budget, which number to trust, which creative to run, when to narrow. One agent collapses two roles into one conversation — you no longer keep a designer and a media buyer on payroll, but the call is always yours.

A real run

We ran this for real: using the product name “AutoWhisper,” we pulled 8 real feature posters + 8 lookbooks across workspaces, built a proper campaign AutoWhisper_Traffic_US_2026-07 with an ad set covering the US + Canada + UK + Australia + New Zealand, made 4 static ads racing against each other, rendered real previews — all PAUSED, all $0. The video was left for manual upload, per Gotcha 3.

Not “it works in theory” — it actually ran. Walk through the four things above and you can reproduce the same result.

Next: Once the ads are running and you’ve got inquiries, turn them into real customers — see the judgment discipline in Turning AutoWhisper Content into Leads with TikTok and the email sequence templates for follow-up.