Cloudflare, continuing its push to enhance AI development, announced a range of new tools and features aimed at accelerating the creation of agentic AI applications during its Developer Week event. Building upon the release of the Agents SDK JavaScript framework in February, which is designed to help developers build AI agents, the company revealed several new updates and capabilities to improve the development and deployment of these agents. As part of this initiative, Cloudflare also launched a dedicated landing page that aggregates all of its agent-related tools in one place for easier access.
A major highlight of Cloudflare’s announcements was the extension of its support for the Model Context Protocol (MCP), an open protocol designed to standardize how AI systems interact with external data sources. According to Rita Kozlov, VP of Product at Cloudflare, MCP is becoming increasingly important as a key interface for large language models (LLMs) to consume APIs and tools, a trend that’s drawing significant interest from the development community. In response to this demand, Cloudflare has expanded its MCP support within the Agents SDK, enabling developers to build remote MCP clients equipped with transport and authentication features for secure interactions with external services.
The updated MCP support also introduces seamless integration with popular authentication platforms such as Stytch, Auth0, and WorkOS’s AuthKit. By incorporating a complete OAuth authentication flow into the Agents SDK, Cloudflare makes it easier for developers to connect their AI agents to remote MCP servers securely, without having to manually develop the authentication process. This integration simplifies the process of linking agents with external data and services while ensuring robust security measures are in place.
Additionally, Cloudflare has taken steps to help developers manage costs related to long-running agentic sessions, which can sometimes experience significant periods of inactivity. With the introduction of the WebSockets Hibernation API in McpAgent, developers can now enable stateful remote MCP servers to “sleep” during idle times, preserving their state while reducing resource consumption. Industry analysts, such as Rachel Stephens from Redmonk, have expressed cautious optimism about MCP’s future, noting that while the technology is still in its early stages, Cloudflare’s growing support for MCP within its Agent SDK signals its increasing role in the development of agentic AI systems.