AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

agents-sdk

Build, debug, or review Cloudflare Agents SDK applications using the agents package.

Original source
cloudflare
License
Apache-2.0
Pinned revision
e44f580f9c4fa3bf268e284ae7706c7954ba31fd

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read cloudflare--agents-sdk at revision e44f580f9c4fa3bf268e284ae7706c7954ba31fd, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 21

references/codemode.md

Codemode (Experimental)

Fetch https://developers.cloudflare.com/agents/tools/codemode/api-reference/index.md for complete documentation.

Codemode lets LLMs write and execute code that orchestrates your tools, instead of calling them one at a time. The LLM gets a single "write code" tool; generated JavaScript runs in an isolated Worker sandbox.

When to Use

| Scenario | Use Codemode? | |----------|---------------| | Single tool call | No — standard tool calling is simpler | | Chained tool calls with logic | Yes | | Conditional logic across tools | Yes | | MCP multi-server workflows | Yes | | Simple Q&A chat | No |

Setup

Wrangler Config

{
  "worker_loaders": [{ "binding": "LOADER" }],
  "compatibility_flags": ["nodejs_compat"]
}

Install

npm install @cloudflare/codemode ai zod

Usage

import { createCodeTool } from "@cloudflare/codemode/ai";
import { DynamicWorkerExecutor } from "@cloudflare/codemode";
import { streamText, tool, convertToModelMessages } from "ai";
import { z } from "zod";

const tools = {
  getWeather: tool({
    description: "Get weather for a location",
    inputSchema: z.object({ location: z.string() }),
    execute: async ({ location }) => `Weather: ${location} 72°F`
  }),
  sendEmail: tool({
    description: "Send an email",
    inputSchema: z.object({ to: z.string(), subject: z.string(), body: z.string() }),
    execute: async ({ to, subject, body }) => `Email sent to ${to}`
  })
};

export class MyAgent extends Agent<Env, State> {
  async onChatMessage() {
    const executor = new DynamicWorkerExecutor({
      loader: this.env.LOADER
    });

    const codemode = createCodeTool({ tools, executor });

    const result = streamText({
      model,
      system: "You are a helpful assistant.",
      messages: await convertToModelMessages(this.messages),
      tools: { codemode }
    });

    return result.toUIMessageStreamResponse();
  }
}

With MCP Tools

const codemode = createCodeTool({
  tools: {
    ...myTools,
    ...this.mcp.getAITools()
  },
  executor
});

How It Works

  1. createCodeTool generates TypeScript type definitions from your tools
  2. The LLM writes an async arrow function calling codemode.toolName(args)
  3. Code runs in an isolated Worker sandbox via DynamicWorkerExecutor
  4. Tool calls route back to the host via Workers RPC
  5. External fetch() is blocked by default — sandbox can only call your tools

Network Isolation

const executor = new DynamicWorkerExecutor({
  loader: env.LOADER,
  globalOutbound: null           // default — fully isolated
  // globalOutbound: env.MY_SERVICE  // route through a Fetcher
});

Limitations

  • Experimental — API may change
  • needsApproval tools execute immediately in sandbox (no approval pause yet)
  • JavaScript execution only
  • Requires worker_loaders binding