Anthropic Ships MCP Tunnels and Self-Hosted Agent Sandboxes
What Happened
Anthropic updated Claude Managed Agents on May 19 with a public-beta self-hosted sandbox option and a research-preview "MCP tunnels" feature, enabling agents to invoke internal Model Context Protocol servers via an encrypted gateway. The release targets enterprise deployments where data residency and internal-tool access are blockers.
My Take
This is the unglamorous plumbing that decides who wins enterprise AI. Sandboxes and tunneling are how you get Claude past the CISO, and Anthropic has been quietly more disciplined about this layer than OpenAI. The flip side: every "agent platform" announcement from here on is really an infrastructure-and-compliance product wearing a model's clothes. Vendors who can't speak fluently to network architects and security reviewers will lose six-figure deals to ones who can.
Read Original Source