Anthropic announced two new infrastructure capabilities at its Code with Claude developer conference in London: self-hosted sandboxes (public beta) and MCP tunnels (research preview). Self-hosted sandboxes let enterprises run agent tool execution on their own infrastructure or via providers like Cloudflare, Daytona, Modal, or Vercel, while keeping the agent loop on Anthropic's side. MCP tunnels act as a lightweight gateway allowing agents to connect to MCP servers inside private networks without public internet exposure. Neither feature requires changes to existing Claude Managed Agents integrations. Early adopters include Clay, Rogo, and DoorDash, each using the capabilities for different enterprise AI workloads.
Table of contents
Anthropic sandboxes, where agents executeMCP tunnels, a lightweight gatewayCustomers clarify Claude Managed Agents use casesRogo banks on Claude Managed AgentsDoorDash orders Modal’s custom container runtimeCore primitives remain the same2 Impressions