Building the Foundation: MCP Servers and AI Tooling
Cloudflare began its journey of integrating AI into its engineering stack by focusing on the development of MCP servers and access layers. These components were essential to supporting agentic AI tools and ensuring their utility across the organization. By pooling expertise from various teams, a specialized unit-dubbed the Internal MCP AgentServer Rollout Squad (iMARS)-was formed. This cross-functional team worked in tandem with the Dev Productivity team, which oversees critical internal systems like CI/CD pipelines and automation frameworks.
The primary goal was to create a scalable and secure foundation for AI-driven development. This required a reevaluation of coding standards, review processes, and onboarding workflows to accommodate AI-driven engineering. The result was a robust infrastructure that now supports thousands of employees and millions of AI requests each month.
Key Metrics Highlighting Adoption and Usage
Within the last 30 days, Cloudflare recorded significant adoption metrics for its AI tools. Approximately 93% of the R&D organization utilized these tools, contributing to a company-wide adoption rate of 60%. In terms of raw usage, 3,683 internal users generated 4.795 million AI requests, demonstrating the systems impact on daily operations.
The adoption of agentic AI tools also translated directly into improved developer velocity. The organization observed a sharp increase in weekly merge requests, climbing from a baseline of 5,600 to over 8,700, with peak weeks reaching nearly 11,000. This reflects the tangible benefits of AI integration in terms of productivity and code deployment.
Infrastructure Optimization through AI Gateway
The deployment of the AI Gateway was a pivotal step in scaling Cloudflares AI capabilities. Over the last month, the gateway processed 24.137 billion tokens and handled 2.018 million requests. This infrastructure enables centralized Large Language Model (LLM) routing and cost tracking, ensuring efficient utilization of computational resources.
The team also implemented Bring Your Own Key (BYOK) functionality and Zero Trust authentication to enhance security and maintain compliance. These features ensure that sensitive data remains protected, even as it flows through complex AI workflows.
Engineer-Facing Tools: OpenCode and Windsurf
Cloudflare developed a suite of engineer-facing tools, including OpenCode and Windsurf, to interact with the AI infrastructure. These tools are designed to be compatible with MCP servers and support both open-source and third-party coding assistants. They provide developers with intuitive interfaces for leveraging AI capabilities in their day-to-day tasks.
By integrating these tools with existing Cloudflare products, the company ensured seamless adoption. Developers can now focus on strategic tasks while repetitive coding processes are automated, further enhancing productivity.
Impact on Organizational Practices
Cloudflares AI engineering stack necessitated a shift in how the organization approaches software development. The company revisited its code review mechanisms, standardization practices, and repository management to align with AI-driven methodologies. This rethinking allowed for better propagation of changes across thousands of repositories, ensuring consistency and reducing redundancies.
The continuous improvement of AI tools and infrastructure has not only increased developer efficiency but also set a new benchmark for organizational agility. With this infrastructure in place, Cloudflare is well-positioned to continue scaling its engineering capabilities.