Meta Ads Agentic Skill

Installation

Prerequisites

You need Node.js installed. It comes with npx built in.

Option 1: npx (All Agents)

npx "@opendirectory.dev/skills" install meta-ads-skill --target claude

Supported --target values: claude, gemini, codex, opencode, anti-gravity, openclaw, hermes

Option 2: Native Plugin (Claude Code Only)

Run these two commands inside your Claude Code terminal:

# Add the OpenDirectory marketplace
/plugin marketplace add Varnan-Tech/opendirectory

# Install a skill directly
/plugin install meta-ads-skill@opendirectory-marketplace

Option 3: Claude Desktop App

Step 1: Download Click Code → Download ZIP on this repo’s GitHub page.

Step 2: Install

1. Open Claude Desktop > Customize > Skills > + > Upload a skill
2. Drop the downloaded zip or extracted folder

Overview

The Meta Ads Skill is a comprehensive, production-ready OpenCode skill designed to give LLMs and AI agents expert-level capabilities to orchestrate the Varnan-Tech Meta Ads MCP Server.

By using this skill, an agent transforms into an Expert Media Buyer. It will know exactly how to safely authenticate, explore ad structures, troubleshoot campaign performance (like CPA spikes), discover new audiences, and format massive Meta APIs JSON payloads into beautiful, readable markdown reports.

---

For Agents: How to Use This Skill Efficiently

This skill is designed using a Progressive Disclosure (Hub-and-Spoke) architecture to maximize context window efficiency:

1. The Hub (SKILL.md): The primary entry point. It provides strict guardrails, safety protocols, and the authentication troubleshooting workflow.
2. The Spokes (references/ & scripts/):
   • When you need to perform a specific task (e.g., investigating a CPA spike), read references/workflows.md for the exact step-by-step orchestration strategy.
   • When presenting data to the user, read references/report_templates.md to strictly follow the required Markdown layout.
   • Data Parsing: Meta Ads JSON responses are massive. Always use the provided scripts/formatters.py to condense raw JSON from get_campaigns, get_adsets, or get_insights into clean markdown tables before reasoning over them.

Strict Agent Guardrails

• Context Protection: ALWAYS default to time_range="last_7d" for insights. ALWAYS use limit=10 for listing campaigns/adsets initially.
• Safety First: NEVER execute state-changing tools (create_campaign, update_campaign, clear_database, reset_database) without explicitly showing the parameters to the user and waiting for their affirmative confirmation.

---

Installation & Setup of the Meta Ads MCP