cloudflare-agents

Build AI agents on Cloudflare Workers by registering tools, integrating MCP protocol support, and connecting LLM providers like OpenAI, Anthropic, or Gemini. Use templates for basic, browser, or complex agent scenarios, then deploy to your Cloudflare environment with bindings like D1, KV, R2, or Durable Objects for state persistence.

Yes, MCP protocol support is built into Cloudflare agents, enabling tool registration and execution across multiple LLM providers. This allows your agents to handle tool use, LLM prompting, RAG, web rendering, and scheduled tasks within Cloudflare's serverless environment.

Cloudflare agents support bindings including D1 for databases, KV for key-value storage, R2 for object storage, and Durable Objects for state persistence. These bindings enable agents to read, write, and maintain state across autonomous workflows and tool executions.

Register tools by defining their schemas and handlers, then execute them through the agent's tool-use mechanism. The agent routes LLM tool calls to registered handlers and returns results back to the LLM, enabling autonomous decision-making and workflow execution.

No. Cloudflare agents run on Cloudflare Workers, which is serverless infrastructure. Deploy your agent code directly to the platform using templates; Cloudflare handles scaling, provisioning, and infrastructure management automatically.

Basic templates provide simple agent setup with tool registration and LLM integration. Browser and complex templates add capabilities like web rendering, WebSocket connections, and multi-step workflows for more sophisticated autonomous agent scenarios.