Cloudflare introduces new tools to accelerate the development and implementation of AI agents, including a first external server for the Model Context Protocol and free access to Durable Objects.
Cloudflare aims to play a prominent role in the development of a global AI ecosystem, and is expanding its offerings accordingly. The company is launching a remote version of Model Context Protocol (MCP). This is an open standard that enables AI agents to perform tasks independently via external services.
Where MCP previously only ran locally, the external server now makes it possible for agents to communicate via the internet with services such as email or calendars, without the need for local infrastructure. This allows developers to build scalable applications more quickly.
The remote MCP server remembers context and ensures consistent user experiences. Thanks to integrations with Auth0 and Stytch, users can securely delegate their permissions to agents.
Free Durable Objects
Cloudflare combines this support with the broader availability of Durable Objects, a technology that brings together computing power and storage in a serverless environment. This was previously only available within paid subscriptions, but is now also accessible through Cloudflare’s free offering.
Durable Objects allow AI agents to maintain context across multiple interactions, such as preferences or previous choices. The system scales automatically with demand and supports millions of simultaneous interactions. Due to the global distribution of Cloudflare’s network, AI agents can, in principle, run close to the user, resulting in lower latency.
Finally, Cloudflare is making Workflows generally available. This tool supports building applications that perform tasks over longer periods, such as booking travel or managing recurring processes. Workflows can automatically restart and remain active for periods ranging from minutes to weeks.
These announcements fit into a series of AI-related innovations launched by Cloudflare. The company also wants to protect AI applications, and introduced updates for its serverless AI platform Workers AI.