Railway is building an agent-first deployment platform designed to let AI coding agents handle the full software development loop — from signup and first deploy to debugging, iteration, and operations. Key features include an agent-optimized CLI with JSON output and reduced interactivity requirements, local and remote MCP server support (with the older npx path deprecated), agent skills that teach agents when and how to use Railway tools, sandboxes for disposable execution environments, a Railway-native Agent that understands the platform's project model, and Guardrails policies (starting with deploy source control) that let teams safely expand what agents are allowed to do. MCP-driven agent usage has grown more than 5x in the past month.
Table of contents
State of Railway: AgentsAll aboard, agents. Sign-Up through first deployment.Home for the AgentsThe CLI as an agent interfaceBuilding a better "Up"MCP - Local and Remote, and when less is more.More than just a computer. Giving agents infrastructure.Railway Agent as an operatorRails, not handcuffsWhere we are135 Impressions