Claude Code AI Agent (MCP) Is My No. 1 Employee (you must do this)
Based on All About AI's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
The onboarding agent automates signup handling by querying a Neon database on Vercel for new records and processing them in order using an ID/progress tracker.
Briefing
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.
Cornell Notes
Claude Code is used to automate two core jobs for the Ninth Brain community: onboarding and marketing. For onboarding, the agent watches a Neon database (on Vercel) for new applications, then creates a unique one-time Discord invite and sends a welcome email through Gmail MCP. For marketing, it runs a research-to-video pipeline: web search and fetch inform a script, a voiceover is generated, scene counts are calculated from voiceover duration, images and clips are produced in a chosen style (e.g., South Park), FFmpeg assembles the video, music is generated and mixed, and the result is uploaded to YouTube. The system logs progress and cleans up files, enabling batch processing and repeatable publishing with limited manual effort.
How does the onboarding workflow turn a new application into a Discord invite and email automatically?
What role does the workflow.md file play in making the agent repeatable and controllable?
How does the marketing workflow keep video visuals aligned with the voiceover length?
What are the main stages of the short-form video pipeline in this setup?
What does the operator do to avoid repeating the same marketing content?
Review Questions
- What mechanisms in the onboarding workflow prevent duplicate processing of signup records?
- Describe how the marketing workflow determines the number of scenes from the voiceover duration.
- Which MCP servers/tools are involved in turning a topic into a published YouTube short, and what happens at the FFmpeg stage?
Key Points
- 1
The onboarding agent automates signup handling by querying a Neon database on Vercel for new records and processing them in order using an ID/progress tracker.
- 2
Each new applicant triggers a unique, one-time Discord invite created via Discord MCP and a welcome email sent via Gmail MCP with the invite link embedded.
- 3
The operator’s time savings come from removing manual steps across database lookup, Discord invite creation, and email sending—turning a multi-minute task into seconds.
- 4
The marketing workflow is structured around a reusable workflow.md that defines research, script, voiceover, scene planning, media generation, editing, and publishing steps.
- 5
Scene counts are calculated dynamically from voiceover length so visuals match audio timing more reliably than fixed templates.
- 6
FFmpeg is used to assemble generated clips, while a music MCP generates background audio that is mixed before adding the Ninth Brain CTA and logo.
- 7
After publishing, the system logs video details and cleans up temporary files to keep future runs efficient and avoid repetition.