Anthropic Ships MCP Tunnels and Self-Hosted Sandboxes for Claude Managed Agents — Enterprise Data Stays On-Prem

9to5Mac / Anthropic·medium signal

Anthropic announced two new security features for Claude Managed Agents on May 19: MCP Tunnels (research preview) lets agents reach MCP servers inside private networks without exposing them to the public internet, requiring no inbound firewall rules; Self-Hosted Sandboxes (public beta) keeps sensitive files and tool execution in customer infrastructure while orchestration stays on Anthropic's servers. Partners include Cloudflare, Daytona, Modal, and Vercel. This directly addresses the enterprise objection that agentic AI requires sending sensitive data to third-party infrastructure.