How to Launch Ads on Facebook and Google Through Your Agents

You can now launch ad campaigns without opening Meta Ads Manager. Without touching Google's cluttered interface. Without hiring a marketer who charges $3,000 a month to do what your agent can do in an afternoon.

The tools are live, the integrations work, and builders are already running campaigns from their terminal. Let me show you exactly how.

What Changed: From Manual to Agentic Ad Management

Three things converged in late 2025: MCP servers now connect agents directly to ad platform APIs; Claude marketing skills - pre-built command libraries - hit production-ready status with 19,000+ stars; and safety-first CLIs like AdKit let agents create drafts without risking live accounts.

The result: you type "Launch a Google Search campaign for my SaaS, $3K/month, targeting engineering managers" - and your agent generates campaigns, ad groups, keywords, and responsive search ads. All paused as drafts for your review. This is vibe marketing. And it's here.

Facebook/Meta Ads via Agents: The Graph API + MCP Setup

The Meta Marketing API¹ (built on the Graph API structure) is the foundation. Every campaign, ad set, and ad is a node you can create, read, and modify programmatically. Your agent speaks to it through MCP.

The Skills That Make This Work

Corey Haines' marketing skills repository² - 19,100+ stars, 281,900+ installs across all skills on skills.sh - includes dedicated ads and ad-creative skills covering Google Ads, Meta, LinkedIn, and Twitter/X campaigns. Install just what you need:

npx skills add coreyhaines31/marketingskills --skill ads ad-creative

The ads skill handles campaign planning and budget allocation. The ad-creative skill generates bulk ad copy and visual concepts. For deeper Meta automation, AgriciDaniel's claude-ads³ provides 250+ checks across 22 sub-skills.

The HeyOz Skills: 10 Slash Commands for Meta Ads

heyoz.com⁴ offers 10 Claude Code skills that collectively save 25-30 hours per week:

The /deploy-ads workflow reads a campaign manifest JSON and bulk-creates campaigns, ad sets, and ads via Graph API v21.0, handling rate limiting, retry logic, and generating deployment reports with direct Ads Manager links.

Advantage+ Shopping Campaigns: Let Meta's AI Optimize

If you're selling products directly, Advantage+ Shopping Campaigns (ASC)⁵ deliver an average 4.52x ROAS compared to 3.70x for manual campaigns - a 22% improvement. ASC handles audience targeting, creative optimization, and budget allocation automatically. Your agent's job is to feed it great creative and monitor the output.

Automated Ad Machine: The Full Pipeline

nocodesaas.io⁶ documented a complete three-layer system for generating and deploying 100+ ad variations:

• Layer 1: Brand File - Colors, fonts, product name, screenshots, key messages
• Layer 2: Creative Matrix - Topics x Personas x Styles = every combination. 7 topics x 5 personas x 4 styles = 140 unique ads
• Layer 3: Upload Pipeline - Uses Meta Graph API v21.0 to upload images, create ad creatives, and organize into campaigns with Campaign Budget Optimization

Google Ads via Agents: Official MCP + Open-Source CLIs

Google released their official Google Ads MCP server⁷ on October 8, 2025. It provides natural language diagnostics and GAQL query execution - but it's read-only for now.

For full campaign creation, you have two paths:

Path A: AdKit CLI (Recommended for Safety)

AdKit⁸ is a safety-first CLI designed specifically for AI agents. Every change is a draft until you approve it:

npx @adkit/cli connect

Then prompt your agent: "Launch a Google Search campaign for my SaaS. $3K/month, ICP is engineering managers at 50-500 person teams."

AdKit outputs: - 4 campaigns split by intent (Brand 10%, Competitor 20%, Solution 50%, Problem 20%) - 11 ad groups, 62 keywords - 11 RSAs with optimized headline/description combos - 2 shared negative keyword lists - All paused as drafts for your review

AdKit also connects to Meta Ads and includes competitor intelligence across 300,000+ real ads.

Path B: google-ads-open-cli (Free, Open Source)

google-ads-open-cli⁹ provides read-only access to the Google Ads API v23 via GAQL. Install it with:

npm install -g google-ads-open-cli
npx skills add Bin-Huang/google-ads-open-cli

Works with Claude Code, Cursor, Codex, and any agent that runs shell commands. While read-only, it's powerful for campaign analysis, performance audits, and feeding data to your agent for optimization decisions.

Path C: Composio Universal CLI

Composio¹⁰ connects coding agents to 1,000+ SaaS applications including Google Ads:

curl -fsSL https://composio.dev/install | bash
composio login
composio connected-accounts link googleads
composio tools list --toolkit googleads

As the Composio team noted: "CLIs are eating MCPs. MCPs for all their merit can be token hungry, slow, and unreliable for complex tool chaining. However, coding agents have become incredibly good at working with CLIs".

The Claude Marketing Skills Repo: 33 Skills, 61 CLI Tools

The coreyhaines31/marketingskills repository is the most comprehensive marketing automation toolkit for Claude Code - 19,100+ stars, 281,900+ installs.

The architecture is smart: a product-marketing skill is the foundation. Every other skill reads it first to understand your product, audience, and positioning before doing anything.

Skill categories include: - Conversion Optimization: cro, signup, onboarding, paywalls - Content & Copy: copywriting (116,100+ installs), cold-email, social (60,700+ installs), video, image - SEO & Discovery: seo-audit (126,400+ installs), ai-seo, programmatic-seo - Paid & Distribution: ads, ad-creative - Strategy: marketing-ideas (140 SaaS ideas), marketing-psychology (85,600+ installs), launch, pricing

Install the full suite:

npx skills add coreyhaines31/marketingskills

Or pick specific skills:

npx skills add coreyhaines31/marketingskills --skill seo-audit copywriting ads

The skills.sh marketplace¹¹ tracks install counts - seo-audit leads with 126,400+ installs, followed by copywriting at 116,100+ and marketing-psychology at 85,600+.

Setting Up Your First Campaign Through Claude Code

Here's the exact sequence:

# Install Claude Code + marketing skills
npm install -g @anthropics/claude-code
npx skills add coreyhaines31/marketingskills --skill ads ad-creative

# Set credentials
export META_ACCESS_TOKEN="your_token"
export META_ACCOUNT_ID="your_account_id"
export GOADS_DEVELOPER_TOKEN="your_token"

Then prompt: "Create a Meta ad campaign for my SaaS. Target indie hackers. Budget $50/day. 3 ad sets, 2 ads each. Test problem-aware vs solution-aware messaging."

The $50 Ad Test: Structure, Targeting, Creative

Before you scale, validate. The $50 ad test framework¹² from the indie hacker community is the fastest way to gauge real demand.

The setup: - $50 Google Ads budget per idea at $1-2 CPC = 25-50 clicks - Target keywords from your market research phase - Landing page with email signup or pre-order option

Success metrics: - Landing page conversion: 3%+ email signups - Ad click-through rate: 2%+ - At least 10 people willing to pre-order

Campaign structure by intent: - Brand (10%) - bid defensively on your name - Competitor (20%) - target alternatives - Solution (50%) - your core keywords - Problem (20%) - problem-aware queries

For Meta Ads, start with Advantage+ Shopping Campaigns - they deliver 22% higher ROAS than manual campaigns. Test 3-5 creative variations per ad set and let ASC handle audience targeting initially.

Reading the Results: What CTR, CPC, and Conversion Tell You

If your CTR is below 1.5%, your creative is the problem. If CTR is good but conversion is low, your landing page or audience is off. If both are good but cost per acquisition exceeds your expected LTV, your pricing or business model needs work.

Common Mistakes and How to Avoid Them

Mistake 1: Launching paid ads before you have LTV data. Paid ads require lifetime value data to optimize. You don't have it in the first 90 days. Build-in-public produces conversion data, audience compounding, and trust signals that paid ads cannot.

Mistake 2: Running all channels simultaneously. The discipline is running 2 at a time, gaining traction, then layering in the next.

Mistake 3: Letting your agent publish without review. Always use --mode draft or a safety CLI like AdKit. Never give your agent direct write access to live campaigns.

Mistake 4: Testing with too small a budget. $50 is the minimum for directional data.

Scaling Up: When to Increase Spend

Scale when you hit these thresholds: - ROAS > 2x on $50 test (you're making money even at small scale) - Consistent conversions for 7+ days - Clear winning creative - one ad performing 2x+ vs others - Landing page converting >3% consistently

Scale by 50% increments every 3-5 days. Don't 10x your budget overnight - the algorithm needs time to adjust. Let your agent monitor daily, pause losers, and reallocate budget to winners using skills like /bleed-check and /rebalance.

The marketing department is now one person with Claude Code and a coffee - building research, landing pages, and ad strategies in under an hour. The barrier between "I have a product" and "I have paying customers" has never been thinner.

But ads are just one piece. Before you spend a dollar, get your GTM gaps assessed at launchbuddy.com - we'll find what's missing in your distribution strategy so every ad dollar converts.

Related Content

• How to Make Demo Videos for Launch with Your AI Agent
• Assess Your GTM Gaps with GetLaunchBuddy
• Zero to First 100 Users: A Distribution Playbook for Vibe-Coded Products

Sources and notes

1. Meta Marketing API: https://developers.facebook.com/docs/marketing-apis
2. Corey Haines' marketing skills repository: https://github.com/coreyhaines31/marketingskills
3. AgriciDaniel's claude-ads: https://github.com/AgriciDaniel/claude-ads
4. heyoz.com: https://heyoz.com/blogs/claude-code-skills-for-meta-ads
5. Advantage+ Shopping Campaigns (ASC): https://www.facebook.com/business/help/449542472823759
6. nocodesaas.io: https://www.nocodesaas.io/p/how-i-built-an-automated-ad-machine
7. official Google Ads MCP server: https://www.malcolmgibb.co.uk/blog/the-google-ads-mcp-why-an-ai-agent-could-run-your-ads-for-you/
8. AdKit: https://adkit.so/features/ads-cli/google
9. google-ads-open-cli: https://github.com/Bin-Huang/google-ads-open-cli
10. Composio: https://composio.dev/toolkits/googleads/framework/cli
11. skills.sh marketplace: https://skills.sh
12. $50 ad test framework: https://www.indiehackers.com/post/i-tested-100-startup-ideas-in-30-days-only-3-were-worth-building-here-s-the-exact-framework-i-used-b91cdf7331