Claude Code Marketing Skills

AI marketing skills for Claude Code, Cursor, Windsurf, and other AI coding tools. Audit SEO, analyze ads, research competitors, qualify leads — all from your terminal.

Free skills need no account. Premium skills connect your real data for $9/mo.

Built by Cogny — AI marketing infrastructure.

curl -sSL https://raw.githubusercontent.com/cognyai/claude-code-marketing-skills/main/install.sh | bash

Or manually copy the skills you want:

git clone https://github.com/cognyai/claude-code-marketing-skills.git
cp -r claude-code-marketing-skills/skills/* .claude/skills/

For Codex, copy them into your Codex skills directory:

git clone https://github.com/cognyai/claude-code-marketing-skills.git
cp -r claude-code-marketing-skills/skills/* "${CODEX_HOME:-$HOME/.codex}/skills/"

These skills work immediately using web search and public data:

These skills connect to your actual ad accounts and email platforms via Cogny's MCP servers:

One person wiring agents into a machine that runs experiments every week: researching the customer, building pages and creative, launching tests, and killing the losers. Inspired by Greg Isenberg's Marketing Engineers playbook. Every agent reads one shared brain in the Cogny context tree. Every campaign carries an ID from the first ad to the closed deal. Every decision lands in a review queue that takes under a minute per item. Guardrails live in the tools: ads are created paused, outreach is drafted and never sent, and scheduled agents get pause-only grants.

CRM: HubSpot is the default (read-only). Salesforce, Pipedrive and Attio map onto the same jobs through the CRM layer in /marketing-engineer.

Beta: these skills are tested live on Cogny Cloud with Google Ads, Meta, X, GA4, GTM, Search Console, Bing, Mailchimp, Get a Newsletter, Stripe and Discord. Other integrations are built from the Cogny MCP proxy source but haven't run against real accounts yet. If something breaks, open an issue.

Start here: /marketing-engineer setup → /marketing-brain → /customer-language. Add the next agent only once you trust the last one.

1. Sign up at cogny.com
2. Connect the accounts you use. Skills pick whichever provider in a category is connected (see Supported integrations)
3. Add Cogny to your .mcp.json:

{
  "mcpServers": {
    "cogny": {
      "type": "http",
      "url": "https://app.cogny.com/mcp"
    }
  }
}

4. Claude Code will open a browser for OAuth login on first use.
5. Run any premium skill — your AI now has direct access to your marketing data.

On Solo, skills record what they find with create_finding. On Cloud, they use tickets (create_ticket) and can also use BigQuery, scripts, automations and image generation. Every skill handles both plans.

Skills are written against jobs ("ad-level spend", "campaign report", "lost deals"), not one vendor. A skill runs on any connected provider in the category and reports each one separately when several are connected.

The routing rules every skill follows (tool names, plan differences, choosing a provider, same-job tables) are in skills/marketing-engineer/references/mcp-routing.md.

Free skills use WebFetch and WebSearch to analyze publicly available data. No accounts or API keys needed.

Premium skills use Cogny's MCP gateway, which federates 60+ integrations through the Cogny MCP proxy. Your AI runs locally (Claude Code, Cursor, Codex, etc.), and Cogny holds the credentials and serves the data.

Your Claude Code ──MCP──> Cogny MCP gateway ──> Cogny MCP proxy ──OAuth/Token──> Google Ads / Meta / LinkedIn / TikTok / Reddit / …
    (local)               app.cogny.com/mcp     (hosted)                         GA4 / Search Console / Semrush / Ahrefs / …
                                                                                 Klaviyo / Mailchimp / Brevo / HubSpot / Shopify / …
                                                                                 (your accounts)

Every connected integration appears as mcp__<Integration>__<tool>, e.g. mcp__Google_Ads__tool_execute_gaql. With the server registered as cogny in .mcp.json, Claude Code shows that as mcp__cogny__mcp__Google_Ads__tool_execute_gaql, which is the form the skills use. Old names like mcp__cogny__google_ads__… are no longer served.

Connected Search Console and Bing Webmaster Tools via Cogny MCP. Ran /seo-audit + /seo-monitor after a site migration.

"Cogny gave Claude Code eyes into our actual search data. The audit caught things we'd never have found manually." — Yi Li, CEO at GrowthHackers

Read the full case study →

Got results? Open an issue and we'll feature you here.

PRs welcome! See individual skill files for the format. Skills should include:

• YAML frontmatter with metadata
• Clear step-by-step instructions
• Specific queries/actions (not vague recommendations)
• Example output format
• Read-only tools in allowed-tools. Write tools (pause, create, send, schedule) prompt the user.

Tool names are checked against the real Cogny MCP surface:

# Regenerate the catalog from the Cogny monorepo (services/mcp-proxy)
python3 scripts/build-mcp-catalog.py ../cogny/services/mcp-proxy/src/servers

# Check every mcp__cogny__… reference in skills/
python3 scripts/check-tool-names.py

MIT