A look at how we built an AI pipeline that discovers Magic: The Gathering news on autopilot, turns it into video content, and publishes it across platforms—while learning from user feedback every week.
Mana Vault is a growing Magic: The Gathering platform that needed to stay relevant in a fast-moving community. MTG news drops constantly—new set reveals, ban list updates, tournament results, price spikes—and the brands that win attention are the ones that post about it first. The problem: manually tracking, creating, and scheduling social content across platforms is a full-time job. Mana Vault needed it to run on autopilot.
Finding MTG news with Hermes cron jobs
The first piece of the pipeline is discovery. We set up Hermes—a lightweight AI orchestration layer—to run scheduled cron jobs that scan the MTG world for noteworthy events. Every few hours, the system checks multiple sources: official Wizards of the Coast announcements, popular subreddit threads, tournament result aggregators, and price tracking APIs.
Each cron job is tuned to a specific signal. One watches for new set or card reveals. Another tracks ban and restricted list changes. A third monitors tournament meta shifts and breakout deck archetypes. When something surfaces, Hermes extracts the key facts—card names, set numbers, date ranges, format implications—and passes them downstream as structured data.
The important detail: these are not mindless RSS dumps. The AI evaluates whether an item is actually newsworthy for Mana Vault's audience before it moves forward. A minor common reprint does not get the same treatment as a Modern-legal rare that reshapes the meta. That filtering step keeps the content pipeline focused on what matters.
Turning data into template videos
Once a piece of news clears the filter, it enters the video template stage. Mana Vault uses a set of pre-designed video templates—intro animations, card spotlight layouts, tournament result brackets, and price movement charts. Each template accepts structured inputs: card image URL, name, rarity, set symbol, and a short headline.
The AI assembles the video by populating the template fields with the extracted data. For a new set reveal, that might mean pulling the key art, listing the highlighted cards, and generating a voiceover script that summarizes the announcement in fifteen seconds. For a ban list update, it formats the affected cards into a clear before-and-after visual.
This is not fully generic AI video generation. It is structured automation: predictable templates filled with real data, producing consistent brand output every time. The team controls the look and feel; AI handles the repetitive assembly.
Publishing through custom MCP servers
The final mile is distribution. Rather than logging into each platform manually, we built custom MCP (Model Context Protocol) servers that connect the pipeline to Mana Vault's social channels. Each server exposes a small set of tools: publish a post, schedule a post, upload a video, or reply to a comment.
When a video is ready, the workflow calls the appropriate MCP server with the content, caption, hashtags, and scheduling parameters. The server handles the platform-specific API details—aspect ratios, character limits, thumbnail requirements—so the upstream workflow stays clean.
This architecture also makes it trivial to add new platforms. When Mana Vault wanted to expand to a new channel, we built a thin MCP server for it and plugged it into the existing pipeline. No rewrite, no new subscription—just a new tool the workflow can call.
Learning from feedback with AI analysis
Publishing is not the end of the loop. Every week, a separate AI workflow reads through the comments, saves, shares, and engagement metrics from every post. It categorizes feedback into themes: what users loved, what confused them, what they want more of, and what fell flat.
The output is a concise weekly digest. Not a fifty-row spreadsheet—a short summary the team can read in three minutes. Examples from a real week might look like:
- Top performer: The Modern meta breakdown video got 3x average saves. Users want more tournament-focused content.
- Underperformer: The price alert post for a common reprint generated confusion. Users thought it was a buy recommendation when it was informational.
- Feature request: Multiple comments asked for deck list links alongside card spotlights.
- Sentiment shift: Positive sentiment dipped on posts about set delays. The team adjusted tone to be more informational and less promotional during uncertainty.
These insights feed directly back into the pipeline. The content strategy adjusts weekly: more tournament coverage, clearer post framing, new template fields for deck links. The system gets smarter not because the model changed, but because the inputs to the template and scheduling logic improved.
What this looks like in practice
The full loop runs with minimal human intervention:
- Cron jobs scan MTG sources every few hours for newsworthy items.
- AI filters and structures the data into template-ready fields.
- Videos assemble automatically from pre-built templates.
- MCP servers publish to each platform on an optimized schedule.
- AI reviews engagement weekly and surfaces actionable feedback.
- The team adjusts templates, filters, and strategy based on the digest.
The result: Mana Vault maintains a consistent, timely social presence without dedicating staff hours to daily content assembly. The team spends their time on strategy and community—not on copy-pasting card images into scheduling tools.
Why this pattern works beyond MTG
The specific domain does not matter as much as the structure. Any business that needs to monitor a fast-moving topic, turn findings into formatted content, distribute it across channels, and learn from engagement can use the same approach. The building blocks—scheduled discovery, template-based content assembly, MCP-based publishing, and feedback-driven iteration—are generic. The value is in wiring them together into one reliable workflow.
How we can help
At CVCraft, we build AI pipelines like this one end to end: discovery, content assembly, multi-channel publishing, and feedback loops. If your team is spending hours on repetitive content tasks that could be automated with structure and guardrails, get in touch. We will map the workflow, design the pipeline, and ship it—with your team owning the result.