Metawork.Studio · Brain

What we built

Metawork Brain is the internal operating system we use to run the agency — and it's built as an MCP server, which means Claude and other AI assistants can directly query our brains. Every client pipeline, meeting transcript, and scenario doc lives in one Git-backed workspace. When we're working on a client project, the AI already knows the context.

Knowledge flows in from everywhere

Brain pulls data from YouTrack (project tasks and timelines), Fathom (meeting transcripts), Looms (video explainers) and our Make scenarios. Everything syncs into structured Markdown files. The Git backend means full version history, easy diffing, and no vendor lock-in.

The MCP architecture is the key innovation. Instead of copy-pasting context into chat windows, Claude can search across all our projects, read docs, and understand relationships between systems. "What did we discuss with Client X about their webhook setup?" — the answer is instant.

Interactive project timeline

Projects display as collapsible rows with color-coded task bars — green for done, other colors by status. Drag the right edge of any task bar to change its due date, which pushes directly to YouTrack via our backend endpoint.

The tricky part was the drag calculation. Our first implementation used raw pixel distances, which broke at different zoom levels. We rewrote it to use day-index math — snapping to calendar days regardless of viewport width. User filter chips let us focus on specific team members; error toasts surface any sync failures immediately.

Lean tools built in a blink

The content editor started as a necessity — we needed to update project summaries without switching to a code editor. It grew into a dual-mode Markdown editor: Toast UI WYSIWYG for visual editing, raw Markdown mode for precision. The floating table toolbar handles the tedious stuff — adding rows, deleting columns, reordering — with single clicks.

Hero image upload, frontmatter editing, tree sidebar navigation — each feature took hours, not days. That's the benefit of owning your own tooling. When we needed README auto-summaries, we wired up the Anthropic API and had AI-generated descriptions flowing the same afternoon.

The router interfaces follow the same philosophy. Small, focused tools that do one thing well. No sprawling dashboards, no feature bloat — just what we actually need to run client work efficiently.