Claude Code AI Agent (MCP) Is My No. 1 Employee (you must do this)

An AI agent built with Claude Code is running a community end-to-end—automatically onboarding new members and producing short-form marketing videos—so the operator can spend minutes instead of hours on repetitive admin and content tasks. The workflow is anchored in a “workflow.md” file that defines step-by-step actions and ties together multiple MCP servers (database, Discord, email, web research, media generation, and YouTube upload). The result is a system that turns new signups into one-time Discord invite links and welcome emails, then uses research-to-video automation to publish engaging shorts that drive traffic back to the community.

The onboarding pipeline starts when someone applies at ninthbrain.com. The agent checks a Neon database hosted on Vercel for new signups, then generates a unique, one-time Discord invite using a Discord MCP. It pulls the applicant’s email from the database and sends a welcome message through a Gmail MCP, embedding the invite link. A progress tracker in the database ensures the agent processes the next unhandled record (the transcript notes skipping IDs 109–111 and then handling 112 next). In practice, the operator demonstrates submitting an application form and then receiving an email confirming selection and providing the invite link. The time savings are framed as substantial: roughly 20–30 seconds for the automated flow versus several minutes of manual work involving database lookup, Gmail, and Discord invite creation.

A second workflow handles marketing automation. It begins with web search and web fetch to research a topic, then generates a script using a pre-defined “shorts formula” tailored to a Gen Z / meme-and-internet-culture audience. The script structure includes elements like hook, escalation, climax, and a call to action, and the agent generates voiceover audio from a chosen voice ID. To keep visuals aligned, the workflow calculates the number of scenes dynamically based on voiceover length (for example, a 30-second voiceover maps to six scenes at five seconds each; shorter durations still get enough scenes to cover the full audio).

Next comes scene planning and image generation in a specified style set—today’s run uses a “South Park style” prompt option—followed by clip assembly. FFmpeg stitches the generated clips together, then a music MCP generates background audio from a prompt (e.g., “chill Bitcoin crypto vibes”), mixes it at a controlled volume, and adds the Ninth Brain call to action and logo. Finally, the agent uploads the finished short to YouTube via the YouTube MCP, logs a summary and URL to avoid repeating the same content, and cleans up temporary files and assets.

The operator acknowledges imperfections—timing between voiceover and visuals can be off—but reports the output is “pretty good” for fully autonomous generation. A prior short is cited as crossing 1,000 views and continuing to climb, suggesting the system is producing usable marketing traction with minimal ongoing effort. The broader takeaway is that Claude Code + MCP can be used beyond coding—turning structured markdown workflows into reliable automation for community operations, content production, and publishing.