Skip to main content

Documentation Index

Fetch the complete documentation index at: https://adadvisor.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

You’ve connected the AdAdvisor MCP server to your AI assistant. Your AI can now read your Meta (Facebook) ad campaigns, run analyses, and make changes. So why does it still need hand-holding to do anything beyond pulling numbers? Because raw tool access isn’t expertise. Your AI knows it can call get_performance, but not when to call it, what to compare against, or what to do with the answer. AdAdvisor Skills close that gap.

​
What is an AdAdvisor Skill?

An AdAdvisor Skill is a packaged Meta ads workflow. One folder containing a SKILL.md instruction file and supporting reference docs. When you install a Skill, your AI assistant loads it automatically the moment a relevant request comes in (for example “audit my account”, “why did my ROAS drop”, “launch a new campaign”) and follows the codified playbook instead of improvising. Think of it as the difference between handing a new media buyer the Meta Ads Manager login (which is what MCP alone does) and handing them the login plus a senior buyer’s training manual covering every common Facebook ads workflow. The MCP is the access. Skills are the brain.
Skills are an open Anthropic standard. They work in Claude.ai (web and desktop), Claude Code, the Anthropic API, and any host that supports the Agent Skills specification. They are not AdAdvisor-specific. We just happen to ship a good set for Meta ads.

​
The eight AdAdvisor Skills

Each skill is a workflow that triggers on specific user phrasing and follows a senior-buyer playbook end to end.

Foundation

Loads on any AdAdvisor task. Account discovery, business context, unit economics, KPI decoding. Every other skill builds on this one.

Audit

Full account audit: pixel integrity, structure, performance baseline, red-flag detection. The first thing a senior buyer does on a new Meta ad account.

Diagnose

Performance investigation. Hook to hold to CTR to CVR funnel diagnosis. Kill, hold, and scale decision rules expressed in multiples of your target CPA.

Launch

End-to-end Meta campaign launch. Objective, structure (ASC vs manual, CBO vs ABO), targeting, creative attachment, activation. Pause-by-default safety.

Scale

Scale proven Facebook ads winners. Vertical vs horizontal, the 20% rule, bid-cap method, target-ROAS portfolio. Eligibility checks before any budget bump.

Creative

Creative strategy. 3-2-2, 5-3-1, and 3-3-3 testing frameworks. Refresh cadence by spend tier. Hook to body to CTA hierarchy.

Targeting

Audience strategy. Broad-first prospecting, lookalike tiers, retargeting funnels, exclusion logic. Post-iOS 14 reality for Meta ads.

Playbooks

Multi-week tactical SOPs. BFCM timeline, new-product launch, lead-gen setup, banned-account recovery, 1kto1k to 10k a day scaling plan.

​
Why use Skills instead of just prompting

You can prompt your AI manually every session (“first read the context, then call get_performance with a 28-day window, compare ROAS to break-even, then…”). It will work, sort of, until the next session when you have to type the same paragraph again. Skills solve four problems at once:

Consistency

Same workflow every time, every session. No more “remember to read the context resource first” reminders.

Discoverability

Skills auto-trigger on natural phrasing. Say “audit my account” and the audit workflow loads. No commands to memorize.

Depth

Each skill bundles reference docs that load on demand. The agent reads the audit checklist when running an audit, the kill-rules reference when killing campaigns.

Expertise

Every threshold, decision rule, and anti-pattern was synthesized from senior practitioners (Foxwell, Tichenor, Faris, Common Thread Collective, Pilothouse). Not guessed by the model.

​
How does a Skill trigger?

You install the Skill once. After that, your AI loads it automatically when you say something that matches its triggers. Example flow: 
You: "Can you take a look at my ad account? Something feels off, CPAs have been creeping up."

Claude: I'll audit the account. Loading adadvisor-audit and adadvisor-diagnose
(since you mentioned rising CPA).

Calling adadvisor:list_ad_accounts to find your account...
Reading adadvisor://account/881523100949350/context...
Calling adadvisor:get_pixel_health...
Calling adadvisor:get_performance with 28-day window...

# Account audit: AdAdvisor Main Account

## TL;DR
Status: 🟡 YELLOW. Three critical issues:
1. Active campaign in permanent learning (budget below threshold).
2. CAPI server-side firing at 2.8%, modeled-conversion inflation likely.
3. Subscribe campaigns at $293 CAC vs $50 implied target.

[... full audit report ...]
Without the skill, Claude would have to figure out what an audit means, in what order, against what thresholds. With the skill, it follows the playbook.

​
Are AdAdvisor Skills free?

Yes. The skills themselves are open source under Apache 2.0 and free to install in any compatible AI host. They require an AdAdvisor account to be useful (since they rely on the MCP server reading your Meta ad data), but the skills package itself costs nothing. Browse the source at github.com/AdAdvisor/skills.

​
How do I install AdAdvisor Skills?

Claude.ai & Claude Desktop

Upload one ZIP per skill via Settings, Capabilities, Skills. Start with the foundation, then add the rest as you need them.

Claude Code

Install the full plugin in two commands. Plugin marketplace handles updates automatically.
Skills require the AdAdvisor MCP server to be connected. Without the MCP, the workflows cannot access your Meta ad data. Set up MCP first if you have not.

​
Frequently asked questions

An AdAdvisor Skill is a packaged Meta ads workflow that your AI assistant loads automatically when you ask a relevant question. Each skill contains a SKILL.md instruction file and supporting reference docs. The eight AdAdvisor skills cover account audits, performance diagnosis, campaign launches, scaling, creative testing, targeting, and tactical playbooks (BFCM, lead-gen, recovery).
The MCP server gives your AI tool access to your Meta ad account. It is the plumbing. Skills tell the AI how to use that access effectively, what order to call the tools in, what thresholds to score performance against, what mistakes to avoid. The MCP is the access. Skills are the brain.
Skills follow the open Anthropic Agent Skills specification, so they work in Claude.ai (web), Claude Desktop, Claude Code, and the Anthropic API. Other MCP-aware hosts that adopt the skills spec will work too. Installation differs by host: Claude.ai and Claude Desktop use a ZIP upload flow, Claude Code uses the plugin marketplace.
No. The foundation skill (adadvisor) is required because every other skill depends on its account discovery and business context logic. The other seven are optional. Install the ones that match how you actually work. Auditing accounts? Install adadvisor-audit. Launching campaigns? Install adadvisor-launch.
Yes. The skills are open source under Apache 2.0. You can install, use, fork, and modify them at no charge. They do require an AdAdvisor account to function (since they read Meta ad data through the AdAdvisor MCP server), but the skills package itself costs nothing.
Each skill declares natural-language phrases in its description field. When you say something matching one of those phrases (for example “audit my account”, “why did my ROAS drop”, “scale this winner”), your AI loads the skill body and follows the workflow. You can also invoke explicitly with slash commands like /adadvisor:adadvisor-audit in Claude Code.
Last modified on May 13, 2026