Anthropic introduces self-hosted sandboxes and MCP tunnels for agents

The company announced two updates for Claude Managed Agents: self-hosted sandboxes and MCP tunnels. These tools let businesses run code and access data within their own secure perimeter, leveraging Anthropic’s cloud logic.
In self-hosted mode, the runtime environment is moved into the client’s infrastructure. This is critical for working with sensitive files or heavy computations. Initially, Cloudflare, Daytona, Modal, and Vercel are supported.
MCP tunnels enable agents to connect to internal databases and APIs without exposing ports to the external internet. The gateway creates an encrypted connection with Anthropic, turning private systems into AI tools.
Sandboxes are available in beta, and tunnel access is provided upon request.
Up next
Bito AI Architect boosts Claude Opus efficiency by 35%
Deep context beats raw model power. A study found Claude Opus performs 35% better with access to the codebase’s knowledge graph. As projects grow to hundreds of files, regular AI agents start to fail. The fix lies not in new model versions, but in feeding project structure correctly through MCP.
Related stories
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

