### Claude Managed Agents Now Include MCP Tunnels and Self-Hosted Sandboxes
Anthropic has recently unveiled two major features for Claude Managed Agents, boosting user oversight concerning security and privacy. These enhancements, MCP tunnels and self-hosted sandboxes, are designed to create a more secure and tailored environment for deploying AI agents.
#### Overview of Claude Managed Agents
Debuted in April, Claude Managed Agents streamline the process of creating and deploying cloud-hosted AI agents. Following the launch of features like dreaming, outcomes, and multi-agent orchestration earlier this month, Anthropic is further extending the functionalities of these agents.
#### MCP Tunnels
The MCP tunnels feature enables users to direct services through a private network, ensuring that Claude Managed Agents can access internal resources without exposing them to the public internet. This feature is especially advantageous for organizations that must uphold stringent security measures.
As per Anthropic, “With MCP tunnels, your agents reach MCP servers inside your private network without exposing them to the public internet. Internal databases, private APIs, knowledge bases, and ticketing systems become tools your agents can call.” This configuration necessitates a lightweight gateway that creates a single outbound connection, avoiding the requirement for inbound firewall rules and public endpoints, while guaranteeing end-to-end encryption of traffic.
At present, MCP tunnels are offered as a limited research preview, requiring users to apply for access prior to implementation.
#### Self-Hosted Sandboxes
The self-hosted sandbox feature enables organizations to set limits for Claude Managed Agents by retaining sensitive files, packages, and services within their own infrastructure or through a managed sandbox provider. While orchestration, context management, and error recovery remain on Anthropic’s infrastructure, tool execution is performed in the user’s defined environment.
Self-hosted sandboxes provide flexibility, allowing the integration of a user’s own sandbox client or making use of services from Anthropic’s partners, such as Cloudflare, Daytona, Modal, and Vercel. This feature is currently in public beta, permitting users to explore its functionalities.
#### Conclusion
These enhancements to Claude Managed Agents demonstrate Anthropic’s dedication to improving security and privacy for users. By launching MCP tunnels and self-hosted sandboxes, Anthropic enables organizations to retain control over their data and infrastructure while utilizing advanced AI functionalities. For further details on these updates, users can visit Anthropic’s official blog.
#### Recent Anthropic News
In related developments, Andrej Karpathy, a research scientist and founding member of OpenAI, has joined Anthropic, indicating a strong commitment to progressing research and development in the field of large language models (LLMs).