Cloudflare has launched Agent Lee, an in-dashboard AI assistant that lets users manage their entire Cloudflare stack through natural language prompts. It can troubleshoot issues, answer account-specific questions, apply configuration changes, and deploy resources across DNS, Workers, R2, SSL/TLS, and more. Technically, it uses a TypeScript-based code generation approach (via Codemode) rather than direct MCP tool calls, with sandboxed execution through Durable Objects acting as credentialed proxies. Write operations require explicit user approval through an elicitation gate, keeping API keys server-side. Currently in beta with ~18,000 daily users and ~250k tool calls per day, it also generates dynamic UI components like charts and tables inline. Future plans include CLI access, mobile support, and proactive monitoring capabilities.
Table of contents
What Agent Lee can doHow we built itBuilt on the same stack you can useGenerative UIMeasuring quality and safetyOur vision aheadTry it outSort: