What are Kelpi Skills?

Runnable skill files for Claude that do the real jobs of running Meta ads: reading a brand, writing ad angles, building campaigns as paused drafts, and auditing the account read-only every day. Each skill ships with its doctrine, the exact prompt, and guardrails.

Do I need an MCP connection to use them?

Only for skills that read or write your ad account. Research and copy skills (buyer language mining, angle writing, copy formulas) work in any Claude chat with no account access. Each skill page states what it needs.

Will connecting an AI to my ad account get me banned?

The known horror stories came from unofficial connectors making burst API calls. These skills are written for connectors that authenticate through official APIs with sane rate limits, and they keep day-to-day access read-only.

Because the prompts are not the hard part. The hard part is showing up every morning to run them. Kelpi is the version that runs on a schedule with guardrails, for $99 a month flat. The skills are the same system, in the open, for people who want to run it by hand.

Kelpi

Kelpi Skills

The Meta Ads Growth OS, as skills you can actually run.

Sixteen skill files for Claude that do the real jobs of paid social: read your brand, write genuinely different angles, launch a $20/day test as a paused draft, and audit the account read-only every morning. Not a PDF about the work. The work.

npx skills add kelpi-ai/meta-ads-skills --all

Installs all 16 into Claude Code, Cursor, Codex, and 60+ other agents via the open skills CLI. Every skill page also has its own one-skill command, the raw SKILL.md, and a chat prompt (guardrails and output bar included) to paste into any Claude conversation.

Run the free audit instead View on GitHub

The rules every skill is built on

Read-only by default

Every skill that touches an ad account analyzes first. Changes are explicit, separate steps you approve. Nothing writes without you.

Honest numbers

No skill invents a metric, rounds a guess into a fact, or claims a change it did not make. If the data is not there, the answer is "the data is not there."

Angles over variations

The offer is fixed; the angle (who + pain + outcome) is the experiment. Five variations of one idea is one test wearing five hats.

Enough spend to judge

No verdict on an angle before it has spent 2-3x your target cost-per-result. Small samples lie politely.

Simple structure

One campaign, one broad ad set, 3-5 genuinely different angles. Low budgets get killed by too much structure, not too little.

Ban-safe

Official API auth, sane rate limits, no burst write loops. The connector horror stories came from tools that did the opposite.

Message & Research

The work that happens before any campaign exists: mine real buyer language, read competitors’ proven angles, sharpen promises, write copy that earns the click.

buyer-language-minerBuyer Language MinerMine reviews, threads, and support messages for the verbatim words buyers use about the problem you solve, organized into pains, triggers, desired outcomes, objections, and alternatives. Use before writing angles or copy; the best hooks are quotes.No account access needed competitor-ad-teardownCompetitor Ad TeardownMine the Meta Ad Library for competitors' PROVEN angles (many variations of one message, longest-running ads) and turn them into angle opportunities for your own offer. Use before writing angles, or when your angles feel stale.No account access needed copy-formulasCopy FormulasWrite or fix ad copy using the classic copywriting formulas (PAS, AIDA, Before-After-Bridge, 4Ps, SCQA, Star-Story-Solution and friends), choosing the right formula for the placement. Use to draft primary text or to diagnose copy that reads fine but does not convert.No account access needed pain-to-promisePain to PromiseTurn a mined buyer pain into a specific, algorithm-aimable promise, and stress-test it with the one-second recognition test. Use to sharpen an angle's promise line before it becomes copy or creative.No account access needed

Brand & Creative

From a pasted website to finished, on-brand ad creative: brand kit, 5-6 genuinely different angles, images you approve before anything ships.

angle-writerAngle WriterWrite 5-6 genuinely different ad angles for one offer (different who, different pain, different promise), each ready to become its own Meta ad. Use when launching, or when testing has killed your current angles.No account access needed brand-readerBrand ReaderRead a website into a working brand kit (offer, voice, colors, fonts, proof) so ads can be written and designed without a creative brief. Use at the start of any campaign, or whenever ads keep coming out off-brand.No account access needed creative-directorCreative DirectorTurn approved angles into finished, on-brand ad images and stage them in the ad account's image library. Use after Angle Writer, before Media Buyer.Writes only with your approval (MCP)

Launch & Targeting

Simple structure that survives small budgets: one campaign, broad targeting, a $20/day test, everything built paused until you say publish.

audience-builderAudience BuilderResearch targeting options honestly (interests, lookalikes, size estimates) and tell you when broad beats all of them. Use before a launch or when delivery seems stuck in the wrong crowd.Writes only with your approval (MCP)lead-gen-builderLead-Gen BuilderBuild Meta native lead forms (drafted for review) with qualifying friction that filters bots and tire-kickers. Use for service businesses that need contact details, not checkouts.Writes only with your approval (MCP)media-buyerMedia BuyerBuild the whole campaign (objective, one broad ad set, a $20/day test, one ad per angle) as a PAUSED draft for review. Use when angles and creatives are approved and it is time to go live.Writes only with your approval (MCP)

Optimize & Protect

Read daily, act rarely: the read-only morning audit, a fatigue rule with no false alarms, and budget rules you approve by name.

budget-pacerBudget PacerPropose budget rules and schedules (spend caps, loser-pauses, winner scaling in 20% steps) as drafts you approve. Use once an angle has proven itself and it is time to scale without breaking it.Writes only with your approval (MCP)daily-auditorDaily AuditorThe read-only morning audit: last 7 complete days vs the prior 7, wasted budget, fading ads, starved tests, pixel problems, one next test. Max 5 recommendations, changes nothing. Use every morning, or right now on any account you suspect is leaking.Reads your account (MCP)fatigue-detectorFatigue DetectorDiagnose whether an ad is actually fatigued (two-condition rule) or you are just impatient, and decide refresh vs leave alone. Use when performance dips or a lead cost suddenly multiplies.Reads your account (MCP)

Measure & Scale

Numbers you can trust before numbers you can scale: honest reporting, commerce plumbing, and tracking that tells the truth.

catalog-commerceCatalog & CommerceSet up product catalogs and product sets so ecom ads can pull live products (dynamic ads) instead of static images. Use for stores with more than a handful of SKUs, after the pixel is verified.Writes only with your approval (MCP)performance-analystPerformance AnalystRead the account like an analyst who is not trying to keep the account: real numbers by angle, platform-ROAS caveats, and the three decisions the data actually supports. Use weekly, or before any scaling or kill decision.Reads your account (MCP)pixel-conversionsPixel & ConversionsVerify the tracking is telling the truth: which events fire, which double-fire or died, and whether Meta's numbers can be reconciled with your store's. Use before launch, after any site change, and whenever dashboards disagree.Reads your account (MCP)

Two ways to run this OS

Run it yourself

npx skills add kelpi-ai/meta-ads-skills --all installs the whole OS; each skill page has a one-skill command, or paste a prompt straight into a Claude chat. Skills that read or write your account use a Meta Ads MCP connection. It is real, it works today, and it costs you the habit of showing up every morning to run it.

Let Kelpi run it daily

The same skills, on a schedule, with guardrails: the audit lands every morning without being asked, launches wait for your approval, and creative gets designed and uploaded. $99/mo flat, 7-day free trial, no card to start, cancel anytime.

Start the free trial

Questions

What are Kelpi Skills?: Runnable skill files for Claude that do the real jobs of running Meta ads: reading a brand, writing ad angles, building campaigns as paused drafts, and auditing the account read-only every day. Each skill ships with its doctrine, the exact prompt, and guardrails.
Do I need an MCP connection to use them?: Only for skills that read or write your ad account. Research and copy skills (buyer language mining, angle writing, copy formulas) work in any Claude chat with no account access. Each skill page states what it needs.
Will connecting an AI to my ad account get me banned?: The known horror stories came from unofficial connectors making burst API calls. These skills are written for connectors that authenticate through official APIs with sane rate limits, and they keep day-to-day access read-only.
Why are these free?: Because the prompts are not the hard part. The hard part is showing up every morning to run them. Kelpi is the version that runs on a schedule with guardrails, for $99 a month flat. The skills are the same system, in the open, for people who want to run it by hand.