Anthropic has taken a step forward in AI agent infrastructure, unveiling new features aimed at enhancing security and functionality at its recent developer conference in London. Among the announcements was the introduction of self-hosted sandboxes and a research preview of MCP tunnels, both designed to meet the current demands of AI software engineers.
Enhancements in Security and Control
The self-hosted sandboxes enable organizations to run AI agents in a controlled environment, ensuring internal networks remain secure from potentially harmful scripts. Traditionally, sandboxes have served for isolated testing and code experimentation. However, in the context of AI, they are crucial for protecting against rogue scripts generated by agents. This setup allows companies to maintain data privacy and runtime control, as agents can now operate on their own infrastructure or that of managed service providers like Cloudflare and Vercel.
These developments are particularly relevant given the rising security and compliance concerns in the AI field.
MCP Tunnels: A New Connectivity Protocol
Alongside the sandboxes, Anthropic introduced MCP tunnels, a lightweight gateway that ensures a secure connection for agents communicating with MCP servers within a private network. This advancement reflects Anthropic’s commitment to establishing stable interconnectivity protocols for agentic software development. The MCP tunnels facilitate a single outbound connection, managed through the workspace settings in the Claude Console, adding an extra layer of security by keeping these connections off the public internet.
MCP's adoption by organizations such as the Linux Foundation has highlighted its growing significance in software development. Anthropic's improvements to MCP tunnels indicate a shift towards more secure, managed connectivity solutions essential for businesses deploying AI technologies.
Implications for the AI Landscape
The launch of these features comes at a time when the demand for secure and efficient AI infrastructure is at an all-time high. As companies increasingly depend on AI to drive productivity and innovation, the tools that support this must also evolve to address security, compliance, and operational efficiency.
One executive stated, “Claude Managed Agents let us replicate the power of a local agent with the reliability, versioning, and background execution of a cloud agent.” This encapsulates the goal of delivering a seamless user experience while maintaining necessary safeguards.
Additionally, with Anthropic’s sandboxes, there is a strong emphasis on using best-in-class infrastructure to enhance the AI platform. “Running it with our sandboxes, like Daytona, gives us control over the filesystem, so we can mount external file stores and install packages on the fly,” the executive noted, underscoring the flexibility these tools provide.
Looking Ahead
As Anthropic continues to refine its offerings, the implications for the broader AI market are evident. The introduction of self-hosted sandboxes and MCP tunnels not only addresses immediate needs but also sets a precedent for how AI infrastructure can evolve to prioritize security and compliance. This is vital as the industry navigates a landscape where trust and reliability are becoming key competitive factors.
Anthropic’s recent announcements mark a significant shift in the development of AI agents, indicating a future where security and operational efficiency take center stage. With the rapid growth of AI applications across various sectors, these innovations may well establish the standard for how businesses integrate and manage AI technologies moving forward.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
