Anthropic has introduced two significant updates to its Claude Managed Agents, focusing on bolstering user control over privacy and security. These enhancements include the introduction of MCP tunnels and self-hosted sandboxes.
MCP Tunnels: Secure Private Network Access
The newly unveiled MCP tunnels enable Claude Managed Agents to access services within a private network without exposing them to the public internet. This feature allows agents to interact with internal databases, private APIs, knowledge bases, and ticketing systems securely. By deploying a lightweight gateway that establishes a single outbound connection, users can avoid inbound firewall rules and public endpoints, ensuring end-to-end encrypted traffic. Currently, MCP tunnels are available as a limited research preview, requiring users to request access.
Self-Hosted Sandboxes: Enhanced Execution Control
In addition, Anthropic has introduced self-hosted sandboxes, allowing users to maintain sensitive files, packages, and services within their own infrastructure or through a managed sandbox provider. While the agent loop responsible for orchestration, context management, and error recovery remains on Anthropic’s infrastructure, tool execution occurs in the user’s configured environment. This feature supports integration with sandbox clients from partners such as Cloudflare, Daytona, Modal, and Vercel. Self-hosted sandboxes are currently in public beta.
These updates underscore Anthropic’s commitment to providing users with greater control over the security and privacy of their AI agents. As AI integration deepens within enterprise environments, such features become crucial in ensuring data protection and compliance with organizational security protocols.
Source: 9to5Mac
