I'm the founder of Scalably. I build and run multi-tenant Claude agent platforms in production, inside the operations of real companies across Southeast Europe. Not demos, not pilots: agents that ship finished work every day.
My focus is the unglamorous part nobody writes about: what it actually takes to run Claude agents in production. Per-tenant isolation, MCP tool integration, eval-driven model routing, persistent memory, and the failure modes you only hit at real volume. I write about that here because almost everyone publishing about the Claude Agent SDK and MCP is building toy examples. I'm running the real thing.
What I work on
Scalably's platform is built entirely on the Claude Agent SDK. It runs across multiple channels (Slack, Telegram, WhatsApp, web), exposes a governed catalog of MCP servers and permissioned skills to each tenant, and routes between Claude models by measured task performance rather than preference. I author MCP servers as production artifacts, including open-source ones.