The AI engineering stack we built internally — on the platform we ship

Cloudflare spent 11 months building an internal AI engineering stack on its own platform, covering MCP servers, auth, routing, inference, review, and sandboxing.

The stack uses Cloudflare products like Access, AI Gateway, Workers AI, Agents SDK, and Sandbox SDK to support secure agentic workflows.

Adoption is now broad: 93% of R&D and 60% of employees use AI coding tools, including OpenCode and Windsurf.

Cloudflare says the platform has increased merge request volume and now routes millions of AI requests per month.