Anthropic has made a notable advancement in AI infrastructure with the launch of self-hosted sandboxes and MCP tunnels for its Claude Managed Agents. These updates, announced at the company's inaugural developer conference in London, aim to improve security and operational control for organizations creating AI applications.
Enhancing Security with Self-Hosted Sandboxes
Self-hosted sandboxes enable AI agents to function in a controlled environment, which is essential for isolating code experimentation and safeguarding company networks. Traditionally, sandboxes are utilized to securely test code, and in the realm of AI, they prevent unauthorized scripts from infiltrating broader networks. Anthropic's strategy places data privacy compliance and security firmly in the hands of the organizations utilizing its infrastructure.
In these sandboxes, agents perform tasks either on the customer's infrastructure or through managed service providers like Cloudflare and Vercel. While execution takes place in the customer’s environment, the continuous agent loop remains under Anthropic's management, ensuring effective control and oversight.
Introducing MCP Tunnels for Secure Connectivity
In addition to the sandboxes, Anthropic unveiled MCP tunnels, a lightweight gateway designed to improve interconnectivity for AI agents. This feature enables agents to connect to private MCP servers without being exposed to the public Internet, significantly boosting security. As a near-standard protocol for agentic software development, MCP tunnels facilitate a single outbound connection managed through workspace settings in the Claude Console, simplifying the integration process for system administrators.
Anthropic emphasizes that both self-hosted sandboxes and MCP tunnels do not require changes to existing Claude Managed Agents integrations. Transitioning between Anthropic's cloud infrastructure and customer infrastructures can be accomplished through simple configuration changes, streamlining the process for developers.
Implications for AI Development
The rollout of these features highlights a growing trend in the AI industry toward enhancing security and compliance in AI development. The ability to mount external file stores and dynamically install packages within the sandboxes provides developers with the flexibility to optimize their AI applications. As one company representative pointed out, "Running it with our sandboxes… gives us control over the filesystem."
The opportunity to utilize stable infrastructure while concentrating on the key aspects of AI product development—tools, data, and user experience—sets a promising standard for financial AI platforms and beyond. Anthropic's clear objective is to offer a highly efficient pathway to production while ensuring total control, scalability, and reliability in AI applications.
Looking Ahead
As the demand for AI solutions continues to rise, securely integrating AI agents into existing infrastructures will be essential. Anthropic’s recent developments not only tackle current challenges faced by AI software engineers but also position the company to play a significant role in shaping the future of AI infrastructure. The enhancements in security and operational control are likely to resonate well within industries that prioritize data protection and compliance, such as finance and healthcare, paving the way for wider adoption of AI technologies in secure environments.
