Two tools, two jobs: let Claude's computer-use agent browse the public Meta Ad Library for you, then use AutoAdy to ground what you find in your own account and build from it.
One honest note up front, because most "AI competitor research" pitches blur this. AutoAdy does not read the Meta Ad Library, and neither does any other ads tool in a way you can rely on. Meta's Ad Library API only returns political and issue ads outside the EU, with no spend or performance data, so you can't pull a typical commercial competitor's ads through it. What you can do is open the public Ad Library website, which shows everyone's active ads to anyone with a browser. That's a job for a browser agent, and Claude's computer-use mode (sometimes run as Cowork) handles it: it navigates, scrolls, screenshots, and reads the ads with vision.
So this playbook splits the work. Claude's browser agent does the competitor side, opening the Ad Library and reading a rival's live ads. AutoAdy does your side, telling you what already converts in your own account and turning the findings into creatives. Connect both and the whole loop fits in one session.
Key takeaways
- AutoAdy can't read the Ad Library, and the public API won't return most commercial competitor ads. A browser agent reading the public Ad Library website is the honest way in.
- A competitor ad that's been running for months is a reasonable hint that it works, not proof. The public library shows each ad's start date; you never see their spend.
- Ground every competitor idea against your own winners with
get_winning_frameworksbefore you spend. What converts for their pixel may not convert for yours.- The session is only worth running if it ends in a brief.
Setup: connect two things (5 minutes)
- AutoAdy MCP, for your own-account tools. In AutoAdy open Settings → API Keys, create a personal key, then add it as a custom connector in Claude under Settings → Connectors.
- Claude with computer-use enabled, so it can drive a browser to the Ad Library. This is a Claude capability, separate from AutoAdy, and it needs computer-use or the Cowork desktop agent turned on.
- Confirm both work: ask AutoAdy "which account am I connected to?", and ask Claude to open the Meta Ad Library and report what it sees.
get_winning_frameworks and ad teardowns work on every AutoAdy plan, including free. Generating creatives with url_to_ads is a Pro capability.
Use case 1: Pull the competitor's live ads (browser agent)
This step runs on Claude's computer-use, not on AutoAdy. It's slower than an API call and worth watching, but it sees exactly the ads you would if you opened the library yourself.
The prompt:
Open the public Meta Ad Library, search for [COMPETITOR], and filter to
[COUNTRY] with active ads only. Scroll through their results and, for each
ad, capture: the hook, the body angle, the offer, the CTA, the format
(UGC / static / video / carousel), and the date it started running.
List the ads that have been live the longest first.
Long-running ads go to the top for a reason: advertisers rarely keep money-losing ads live for months. You can't see their spend or results, so treat run time as a hint, not a verdict.
Use case 2: Find the open angle
Now Claude reasons over what it just read. You're looking for the trigger the whole category has ignored, because that's usually where attention is cheapest.
The prompt:
From the ads you collected, group them by psychological trigger:
curiosity, fear of loss, authority, social proof, urgency, status.
Tell me which triggers the category leans on and which one nobody is
using. I want the angle that's wide open.
Most categories settle into two or three safe angles, often a discount play and an authority play, and leave a whole emotional register untouched. Beating a saturated angle with a slightly better version rarely works. The first credible ad in the empty register usually does.
Use case 3: Ground it in your own winners (AutoAdy)
A competitor's angle tells you what's in the market. It can't tell you what works for your audience, offer, and pixel. Before you spend, check the idea against what already converts for you. This is where AutoAdy earns its place.
The prompt:
Across my connected ad accounts, show me the winning frameworks: which
audience, copy length, hook, angle, and visual style produce my best
blended cost per lead, only where there's enough spend to trust it.
Then tell me which competitor angle overlaps with something already
proven for me.
This calls get_winning_frameworks, which works on every plan. Pair it with a teardown of one of your own top ads:
Give me the full teardown of my ad [AD ID]: the copy (primary text,
headline, description, CTA), the targeting (age, gender, locations,
interests, custom audiences, placements), the budget, and the objective,
so my next creative inherits what already works for me.
That second prompt calls analyze_winning_ad. It pulls the details you'd otherwise have to dig for, so the next creative starts from a proven base instead of a blank page.
Use case 4: Brief it, then build (AutoAdy)
End the session by forcing a brief, then generate the first batch from your own page.
The prompt:
Combine the open competitor angle with my proven framework into one
creative brief for the next test: the angle and why it's the gap, the
hook and framework to build on, the audience based on what converts for
me, and 3 concepts. Then generate a first batch from my landing page
[URL] built around that angle.
The brief is Claude reasoning. The generation calls url_to_ads (Pro), which builds from your own URL and your brand DNA, with every creative staged paused for you to review.
How the steps stack
| Stage | Runs on | What it does |
|---|---|---|
| Pull competitor ads | Claude computer-use (Cowork) | Browses the public Ad Library and reads a rival's live ads |
| Find the open angle | Claude reasoning | Groups the ads by trigger and names the gap |
| Ground in your winners | AutoAdy (get_winning_frameworks, analyze_winning_ad) | Ranks what actually converts in your own account |
| Brief and build | AutoAdy (url_to_ads) | Turns the gap plus your proven framework into staged creatives |
AutoAdy's edge here isn't spying on competitors. Plenty of tools claim that, and most can't actually pull the ads. Its edge is knowing what converts in your account and turning a brief into creatives quickly. Pair that with a browser agent for the reconnaissance and you get the whole loop, without anyone pretending the Ad Library API does something it doesn't.
What this means
Competitor research is two separate jobs, and using the right tool for each is what keeps it honest and fast. A browser agent can read the public Ad Library the same way you can, only without the three hours of scrolling. AutoAdy can tell you which of those ideas stands a chance in your account and build the test. Run them back to back and "what's the gap, and what do I build?" becomes a single sitting instead of a half-day project.