Cloudflare Agents

Cloudflare Agents provides a comprehensive platform for building, deploying, and scaling AI agents on Cloudflare's global edge network. The platform combines Cloudflare Workers with Durable Objects to enable stateful, long-running AI agents that can maintain context, communicate in real-time, and integrate with external tools and APIs.

• Stateful Agent Architecture uses Durable Objects to maintain persistent state across interactions, allowing agents to remember context, track conversations, and manage complex multi-step workflows without external databases.

• Real-time Communication supports WebSocket connections for bidirectional streaming, enabling agents to push updates, handle interruptions, and maintain live conversations with users.

• Tool Integration allows agents to call external APIs, execute functions, and interact with databases through a flexible tool-calling system that supports both synchronous and asynchronous operations.

• Human-in-the-Loop Support enables agents to pause execution and wait for human approval or input before proceeding with sensitive operations, providing control over automated workflows.

• Scheduled Tasks lets agents schedule future actions using built-in cron-like functionality, enabling autonomous background processing and time-based triggers.

• Model Context Protocol (MCP) integration allows agents to connect to MCP servers for extended capabilities, accessing additional tools and data sources through a standardized protocol.

• Multi-Model Support works with various AI providers including OpenAI, Anthropic, Google AI, and others through the AI SDK, giving developers flexibility in choosing their preferred language models.

• Browser Rendering provides headless browser capabilities for agents that need to interact with web pages, scrape content, or automate browser-based tasks.

To get started, developers install the Agents SDK via npm and create a new agent class extending the base Agent class. The platform handles deployment, scaling, and state management automatically. Agents can be tested locally using Wrangler and deployed globally with a single command. The SDK provides TypeScript support with full type safety for agent state, tools, and message handling.