MCP clients like Cursor, Windsurf, and Claude can now connect directly with Cloudflare’s infrastructure, enabling account automation, information processing, and more. This is thanks to the launch of Cloudflare’s remote Model Context Protocol (MCP) server suite. The company has initially made public 13 such offerings, each one providing a specific service that can be provided through natural language.
The remote MCP servers were introduced in March, providing developers an alternative means of building with AI models and agents. At the time, Cloudflare notes that MCP was limited to operating locally on a server instead of through a pre-existing option. Now, developers can utilize MCP through a web or mobile app.
This release coincides with Cloudflare’s partnership with Anthropic. Last week, the AI company announced new ways to pair third-party apps and tools with its popular AI chatbot.
“The way users want to interact with Cloudflare services are quickly changing now with the help of large language models (LLMs),” Rita Kozlov, Cloudflare’s vice president of developers and AI, tells me in an email. “The Cloudflare MCP servers allows agents to not only query a user’s account and process information, but also take action on a user’s account. This enables any user to understand and interact with Cloudflare’s offering without having to be an expert on every API, feature, or documentation page, by handing off complexity to the LLM.”
Thirteen might seem like an odd number of servers to start with, but it’s based on the number of use cases Cloudflare employees identified during brainstorming. There’s a Cloudflare Documentation server that shows reference information from the company’s Developer Documentation.
Other servers include a Workers Bindings server (build Workers applications with storage, AI and compute primitives), Container server (quickly stand up a sandbox development environment), Radar server (retrieve global Internet traffic insights, trends, web scans, and other utilities), AI Gateway server (search logs and get prompt and response details), DNS Analytics server (optimize DNS performance and debugging issues based on current configurations), and a Cloudflare One CASB server (rapidly identify SaaS application security misconfigurations).
“We plan to continue building out MCP servers to support Cloudflare’s services to allow any user to have a better understanding of their website, how they can improve the performance and security of their application,” Kozlov states. She adds that support for the company’s Zero Trust products and developer platform is planned.
Atlassian, PayPal, Sentry, and Webflow are some of Cloudflare’s customers now accessing these remote MCP servers. The time from planning to launch is low, too—Kozlov reveals that MCP servers could be built “in less than 24 hours.” As she explains, if an MCP client has support for remote MCP servers, the client should have a way to accept “the server URL directly within its interface.” For example, with Claude.ai, developers can go into their settings, add the URL of the MCP server into the “Integration” screen, authenticate with Cloudflare, and specify the tools they want Claude.ai to be able to call.
“MCP makes it easy for users to interact with a product or service, even if they don’t have a lot of prior knowledge with it—without having to read product documentation or take time to learn the technical side of things. The faster a user can experience the value of your product, the more likely they will be to enjoy, upgrade, and continue to use your product,” Kozlov points out. “MCP has the potential to create all sorts of connections we haven’t even thought of yet. As it catches on, it has the potential to be the new mode of interaction with apps and tools, just like mobile was.”
She emphasizes that just like how Cloudflare helped make the internet fast, reliable, and secure, it’s looking to do the same for AI. “A lot of these AI tools are going to need to gain our trust as users—and the faster, safer, and more performant these experiences are, the more natural they’re going to feel, and we’ll build new habits around them.”
Featured Image: An AI-generated image of a computer server room. Image credit: Adobe Firefly
Leave a Reply
You must be logged in to post a comment.