How I use Claude Code + MCPs to run my marketing

Cody Schneider defines a real marketing agent as requiring three components: solving the data problem, giving the agent unified visibility across the entire pipeline, and having the agent autonomously make decisions on a recurring cadence with a thinking loop. He distinguishes this from what most people calling themselves marketing agent builders are actually running, which he says are linear automation workflows in tools like Zapier or N8n rather than truly autonomous systems.

The infrastructure required to run a genuine marketing agent includes a data pipeline, a data warehouse, and cloud hosting for the agent itself. Schneider recommends Airbyte feeding into ClickHouse, both open-source and self-hostable, with the entire setup buildable using Claude Code. The warehouse unifies Facebook ads data, Google Analytics, PostHog, HubSpot, and Stripe so the agent can tie specific ads to actual revenue rather than surface-level metrics.

The production Facebook ads system Schneider describes creates two ad sets per day with five ads per ad set, automatically uploaded to Facebook. Ads run for a two to three day window to gather signal, the worst performers are turned off, and winners enter a pool competing for available budget. The agent handles the full cycle including researching customer pain points via Reddit and Perplexity, generating static and AI avatar UGC creative via tools like Kie AI and HeyGen, publishing via the marketing API, and looping performance data back to improve future creative. Schneider warns that Facebook ad accounts get banned not from using agents but from violating terms of service by spamming the API with bulk data reads, and says the Facebook marketing API should only be used for write operations, with data reads handled through the warehouse pipeline instead.

Facebook's Andromeda algorithm analyzes ad creative including images, text, video scripts, and landing page content to match ads to users, making interest-based targeting no longer necessary. Schneider describes Facebook as currently the best B2B ads channel available, capable of identifying and serving niche B2B ads to as few as roughly ten qualifying people in the US and still delivering qualified leads.

A key failure mode Schneider identifies in these agent loops is entropy, where the agent gets stuck generating the same type of creative over time and performance degrades from day one onward. His solutions include pulling competitor ads from the Facebook ads library and injecting them as new creative DNA, and using transcripts from YouTube videos and podcasts to surface fresh insights. A database of all previously created ads including the JSON prompts sent to generation tools is maintained so the agent can analyze which creative attributes correlate with revenue outcomes.

Schneider identifies WordPress as a blue ocean opportunity because it powers over 43 percent of websites and no one is building AI-first tools for it. His proposed approach is to find existing WordPress plugins with validated demand and no AI component and build the AI-first replacement. Specific examples include replacing Yoast SEO with an agent that writes meta tags, restructures content, and adds internal links automatically; replacing WP Forms with a conversational AI form that qualifies leads; and enhancing WooCommerce with an AI storekeeper that writes product descriptions and abandon cart flows. The business model bundles plugins users currently pay for separately into a single package sold via a token model with a base tier around 29 dollars per month.

The agents described span most major marketing channels. A Google Ads agent can run an entire ad account autonomously. An influencer outreach agent researches influencers in a given category, sends cold emails, negotiates pricing, and flags strong candidates for human review. A cold email agent finds target customer addresses, sends outreach, and manages the inbox through to booking meetings. A TikTok reel farm involves spinning up roughly 10 accounts in the cloud, generating product-related slideshows at scale, and aggregating view counts across accounts. For AI search visibility, a citation-building approach uses cold outreach to get a brand listed on sources already appearing in results for a given category. An email newsletter agent can research podcasts, write and narrate scripts using ElevenLabs voices, and build lead magnets to grow an industry-specific email list. Separate agents can run LinkedIn and Twitter accounts autonomously, and SEO agents handle keyword research and write on-brand articles with a distinct point of view to improve both rankings and readability.