Anthropic is introducing two new features for Claude Managed Agents that give users more control over security and privacy. Claude Managed Agents now support MCP tunnels and self-hosted sandboxes Anthropic unveiled Claude Managed Agents in April, significantly simplifying the work required to create and deploy cloud-hosted AI agents. Earlier this month, Managed Agents went even further with new Dream, Results, and Multi-Agent Orchestration features. Anthropic is now adding two additional new features for May: MCP Tunnels and Self-Hosted Sandboxes. “The sandbox in which an agent runs the tools and services it hits run within the established boundaries of your business, under your security and execution controls,” Anthropic explains. The new MCP tunnels feature allows Claude Managed Agents users to route services through a private network. With MCP tunnels, your agents reach MCP servers within 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 on. A lightweight gateway that you deploy establishes a single outbound connection, no inbound firewall rules, no public endpoints, and end-to-end encrypted traffic. MCP tunnels are a limited research preview, so it is necessary to request access before starting at this time. Meanwhile, the new self-hosted sandbox feature allows you to easily set limits for Claude managed agents. With self-hosted sandboxes, you keep sensitive files, packages, and services in your own infrastructure or with a managed sandbox provider. The agent loop that handles orchestration, context management, and error recovery remains on Anthropic’s infrastructure, while tool execution moves to your own configured environment. Self-hosted sandbox supports bringing your own sandbox client or using one of Anthropic’s partners: Cloudflare, Daytona, Modal, and Vercel. Self-hosted sandboxes are coming as a public beta feature. You can learn more about the two Claude Managed Agents updates here. Recent news from Anthropic: Separately, researcher and founding member of OpenAI Andrej Karpathy joined Anthropic after being independent for some time: Personal Update: I have joined Anthropic. I think the next few years on the LLM frontier will be particularly formative. I am very happy to join the team here and return to R&D. I remain deeply passionate about education and intend to return to work in this field in time. – Andrej Karpathy (@karpathy) May 19, 2026 FTC: We use automatic, revenue-generating affiliate links. More. Post navigation macOS 27: five new Mac features announced next month
