Anthropic Expands Enterprise AI Strategy With Self-Hosted Sandboxes And MCP Tunnels

AI startup and research company Anthropic announced that it has introduced new deployment options for Claude Managed Agents that allow enterprises to run agent workloads inside their own infrastructure while keeping core orchestration on Anthropic’s systems. The rollout includes self-hosted sandboxes, now available in public beta, and MCP tunnels, which are in research preview and require access approval.

The update is aimed at companies that want to use AI agents without moving sensitive files, code, or internal systems into a third-party cloud environment. Under the new setup, Anthropic keeps the agent loop responsible for orchestration, context management, and error recovery, while tool execution takes place in the customer’s controlled environment or with a managed sandbox provider. Anthropic said the model is designed so that enterprise security controls, audit logging, network policies, and runtime restrictions remain in place by default.

The sandbox layer can run on a company’s own infrastructure or through providers such as Cloudflare, Daytona, Modal, and Vercel. Anthropic described the approach as a way to give organizations more control over compute resources, runtime images, and system access, particularly for workloads that require heavy processing such as long-running builds or image generation. The company also said the design allows files and repositories to remain inside the enterprise perimeter.

Anthropic Separates Orchestration From Execution

Anthropic framed the launch as part of a broader push to make agents more practical for enterprise use, especially where privacy, governance, and operational control are central concerns. The company’s architecture effectively separates the “brain” of the agent from the execution layer: planning and coordination remain handled by Anthropic, while actual work is carried out on the customer side.

MCP tunnels extend that model further by allowing agents to reach private internal systems without exposing them to the public internet. Through an encrypted connection established by a lightweight gateway, agents can access databases, private APIs, knowledge bases, and ticketing systems as tools. Anthropic said the setup does not require inbound firewall changes or public endpoints, and that traffic remains encrypted end to end.

The company said MCP tunnels are available in Managed Agents and the Messages API, with administration handled through workspace settings in the Claude Console. The release comes as enterprise AI tools increasingly move toward deployments that combine cloud-based intelligence with customer-controlled execution environments, reflecting a wider industry emphasis on security, compliance, and infrastructure flexibility.

The post Anthropic Expands Enterprise AI Strategy With Self-Hosted Sandboxes And MCP Tunnels appeared first on Metaverse Post.