Cloudflare Speeds AI Agent Development with First Remote MCP Server

Cloudflare, Inc. introduced new solutions designed to speed up AI agent development. With the industry’s first remote Model Context Protocol (MCP) server, general availability of durable Workflows, and a free tier for Durable Objects, Cloudflare makes it faster and easier for developers to build and deploy AI agents — enabling creation in minutes instead of months, while reducing cost and complexity.

AI agent – AI-enabled system that can act autonomously, make decisions, and adapt to changing environments – represent the future of AI. AI agents have the potential to unlock massive productivity gains, and yet businesses are struggling to build agents that deliver real return on investment. Building agents requires access to three core components: AI models for reasoning, workflows for execution, and APIs for access to tools and services. In order to build scalable agentic systems, organizations need access to a platform that can provide each of these components in a scalable, cost-efficient way.

“Cloudflare is the best place to build and scale AI agents. Period. The most innovative companies out there see that agents are the next big frontier in AI, and they’re choosing Cloudflare because we’ve got everything they need to move fast and build at scale on our Workers platform,” said Matthew Prince, co-founder and CEO of Cloudflare. “Cloudflare was built for this moment. First, we built the most interconnected network on the planet. Then, we built a developer platform that took advantage of that network to run code within 50 milliseconds of 95% of everyone online. And, we’re keeping our foot on the gas to give developers the best tools to build the future of agentic AI.”

With this announcement, Cloudflare’s Developer Platform addresses some of the most critical challenges in building AI agents by:

Unlocking smart, autonomous actions with the industry’s first remote MCP server

MCP is a fast-growing open source standard that lets AI agents interact directly with external services. This shifts AI from simply giving instructions to actually completing tasks on a user’s behalf – whether that’s sending an email, booking a meeting, or deploying code changes. Previously, MCP has been limited to running locally on a device, making it accessible to developers and early-adopters but hindering wider, mainstream adoption.

Cloudflare is making it easy to build and deploy remote MCP servers on Cloudflare, so any AI agent can securely connect over the Internet and interact with services, like email, without the need for a locally hosted server. MCP servers built on Cloudflare can retain context, providing a persistent, ongoing experience for each user. And through partnerships with Auth0, Stytch, and WorkOS, Cloudflare is simplifying authentication and authorization that allows users to delegate permissions to agents, making secure agent deployment dramatically simpler.

Building intelligent, contextually aware AI agents with Durable Objects, now on free tier

Previously only available as part of paid plans, developers can now access Durable Objects on Cloudflare’s free tier, expanding broad and democratized access to a critical component for building agents. Durable Objects are a special type of Cloudflare Worker that combine compute with storage, allowing you to build stateful applications in a serverless environment without managing infrastructure. Durable Objects provide the ideal foundation for AI agents that need to maintain context across interactions, such as remembering past preferences or changing behavior based on prior events. Cloudflare’s network ensures Durable Objects scale out to millions of simultaneous customer interactions and can operate agents near the original request, ensuring each customer has a fast, low latency response.

Deploying durable, multi-step applications with Workflows, now generally available

Workflows allows you to build multi-step applications that can automatically retry, persist, and run for minutes, hours, days, or weeks. Workflows is now generally available, providing developers and organizations with a reliable way to build and manage multi-step applications infused with AI. For example, building an agent Workflow to book a trip would require searching for flights in a price range, which would require a persistent search over a certain predetermined time span. Then, once the flights have been found, an agent would purchase the flights with traveler information and a credit card. And finally, send confirmation to each traveler in the party.

Paying only for what you use, for the most cost-efficient AI deployment

AI inference is hard to predict and inconsistent in nature, unlike training, because it relies on human behavior, including the time of day and what action a person wants to take. With traditional hyperscalers, this requires organizations to prepare and provision for the highest level of capacity they can expect, even if that level only happens at peak times. Cloudflare’s serverless platform automatically scales inference and AI agent resources based on demand, from zero to global scale in milliseconds. This ensures organizations only pay for what they use, dramatically reducing costs compared to traditional cloud deployments that require constant provisioning.

“Cloudflare offers a developer-friendly ecosystem for creating AI agents that includes a free tier for Durable Objects and serverless options for AI inference,” explains Kate Holterhoff, senior analyst at RedMonk. “These low-cost, easy-to-use options could empower more businesses to adopt and experiment with agentic AI.”

Cloudflare has been a leader in making AI inference accessible, breaking down barriers that have kept AI out of reach for most businesses. Cloudflare has GPUs deployed across more than 190 cities globally, bringing AI as close to the user as possible for low latency experiences.

Learn more about Cloudflare’s remote MCP server for fast, easy AI agent development at the website here.

Orca Enables Universal GenAI Access with Model Context Protocol