Meta ads for AI agents.
Find the ads that pay, and launch the next one paused. Your agent reads what each creative bought, briefs the next test, and spends nothing until you say so.
✓Accounts are opened by hand during early access.
- meta_get_evidence_health · synced 4h ago · 2 videos read from poster only
- meta_get_history_winners · 90d · founder-led UGC beat its spend share
- meta_suggest_next · test the founder hook on the cold persona
- meta_launch_ads · 1 campaign · 3 ads · PAUSED
- meta_activate · waiting for you to say go
87
MCP tools
Every dashboard action, callable by your agent.
41
Read-only tools
Your agent can look at everything before it touches anything.
PAUSED
Every launch
Campaigns, ad sets and ads are created paused. Always.
0
AI credits per agent call
Your agent runs on your own Claude or ChatGPT plan.
[ 01 / 05 ]Launches land paused
Your agent builds the launch. You turn it on.
Your agent can research, brief, draft and build a whole campaign. Spending money is a separate step, and it's yours.
- meta_launch_ads creates the campaign, ad set and ads PAUSED, ready to review in Ads Manager.
- Turning a launch on is its own call, and your agent asks you first, every time.
- Advantage+ creative enhancements are opted out, so Meta never rewrites the copy you approved.
- Budget and targeting changes are flagged as money-moving and wait for your go-ahead.
[ 02 / 05 ]For your agent
Connect your agent in one command.
No key to paste for Claude Code, Cursor, Codex or claude.ai. They use the same URL and sign in through the browser. Headless agents send an rsk_ key a workspace admin creates.
Then ask — "Check the evidence health, then tell me which creatives actually worked."
claude mcp add --transport http winandscale https://winandscale.com/api/mcp[ 03 / 05 ]What a creative is
It reads the ad, not the ad name.
What a creative is — its angle, persona, offer, hook, format and talent — is stored against the ad, read by a classifier and curated by you. It works on ads that predate any naming scheme.
Values on every creative
Angle, persona, awareness, offer, pain, format, hook, talent and visual.
Videos read in full
The hook, talent and format of the first seconds, not a guess from one still.
Traits that over-delivered
Which Values bought more than their share of spend, over 30, 90 or 180 days.
Lifetime history per ad
Every ad's lifetime totals, beside the daily warehouse.
[ 04 / 05 ]The Test loop
Every test keeps the question it asked.
A next test becomes a Test when you commit to it. Its question and the evidence behind it are frozen, so the answer can't drift to fit the result.
Suggest
The next test, with its basis, its limits, and what would reverse it.
Commit
The question and the evidence it was decided on are frozen.
Launch
Drafts and ads are linked to the Test, and launched paused.
Conclude
Observations are appended, never rewritten. Low delivery is inconclusive, not a loss.
Honest about what it can't see
- Starts every review with evidence health — stale sync, coverage gaps, untagged spend
- Compares funnels on their own goal, never on Meta's Results column
- Names the outcomes Meta can't see, like qualified leads and margin
- Observational, and says so — Meta chose who saw each ad
[ 05 / 05 ]What your agent does
From what worked to what's next.
It can read everything. Anything that spends waits for you.
Find what worked
Winners, the traits behind them, and the tests they suggest.
Research the customer
Turn verbatims into personas and angles, every line traced to a quote.
Read the Ad Library
Save reference ads from the Meta Ad Library to a swipe file.
Draft creative
Read a reference ad into a recipe and generate image drafts from it.
Launch paused
Campaign, ad set and ads built from approved drafts, created paused.
Keep the record straight
Tag creatives, merge duplicate Values, and conclude Tests.
Questions people ask us
Let your agent build the next launch.
Request access, connect your agent, and start with a read-only look at what worked.