Anthropic Launches Self-Hosted Sandboxes and Encrypted MCP Tunnels for Enterprise AI Agents
Summary
Anthropic launches self-hosted sandboxes and encrypted MCP tunnels for enterprise AI agents, letting companies run Claude-powered agents securely within their own infrastructure while keeping sensitive data off the public internet.
Key Points
- Claude Managed Agents now supports self-hosted sandboxes, allowing enterprises to keep agent tool execution within their own infrastructure or with managed providers like Cloudflare, Daytona, Modal, or Vercel, while Anthropic handles the agent orchestration loop.
- MCP tunnels are now available in research preview, enabling agents to securely connect to private MCP servers, internal databases, and APIs without exposing them to the public internet, using a lightweight gateway with end-to-end encrypted traffic.
- Early adopters including Clay, Rogo, and Amplitude are already building production agents on these new capabilities, citing benefits like tighter security controls, faster deployment, and the ability to handle proprietary data within trusted infrastructure.