Claude Managed Agents: self-hosted sandboxes and MCP tunnels

Anthropic has released two updates for Claude Managed Agents: self-hosted sandboxes and MCP tunnels.

The core challenge is that enterprises need agents to execute code and access internal services without exposing sensitive data or violating security policies.

Previously, tool execution lived entirely on Anthropic‘s infrastructure, which limited control over compute and network boundaries.

These updates let organizations keep agent execution within their own perimeter while still benefiting from Anthropic‘s orchestration layer.