I build the infrastructure that lets AI act inside a real stack: custom MCP servers, self-hosted automation, persistent memory. All of it under my own control.
MCP servers, n8n workflows, and a vault as the brain.
The infrastructure the rest of my work runs on. Custom MCP servers connect Claude Code to external APIs, n8n handles event-driven automation on my own VPS, and an Obsidian vault gives agents persistent memory. TypeScript and webhooks, no low-code platform underneath.
An Obsidian vault that Claude can read, query, and write to.
The knowledge layer underneath everything. Decisions, learnings, and skills live as atomic Markdown files with frontmatter, and Claude searches them semantically through a custom MCP tool before it answers.
Custom MCP servers that turn any tool into a typed surface for Claude.
Multiple MCP servers, some built from scratch (Make.com, the Second Brain vault, a few Notion surfaces) and others wired as clients (Asana, Linear, Figma, Stitch, Notebook LM, Excalidraw). Each one exposes its tools through Zod-validated schemas, so Claude calls them like typed functions.