Anthropic debuts MCP tunnels and self-hosted sandboxes to lock down AI agent infrastructure

Anthropic launched a public beta of self-hosted sandboxes and a research preview of MCP tunnels for Claude Managed Agents.

The new setup lets tool execution run on customer or partner infrastructure instead of Anthropic’s public environment.

MCP tunnels also allow private MCP servers to be reached without exposing them to the public internet.

The update gives enterprises more control over security, privacy, and compliance while keeping Claude integrations largely unchanged.