Anthropic Adds Self-Hosted Sandboxes and MCP Tunnels to Bolster Claude Agent Security

When enterprises deploy Claude Agents, the models need to access code, databases and internal tools, potentially taking execution environments and sensitive data beyond existing security boundaries. Anthropic is giving customers control over isolation and connectivity, allowing highly regulated industries such as finance and healthcare to deploy agents on their own infrastructure while limiting the potential impact of security incidents.

At its 2026 developer conference in London, Anthropic introduced self-hosted sandboxes and MCP Tunnels for Claude Managed Agents. The sandboxes keep tool calls within a company’s own environment, while MCP Tunnels connect to internal data through encrypted channels. As of July 19, Anthropic had not disclosed pricing, a formal launch date or supported regions.