Developer Platform Updates & Release Notes

The Agents SDK now includes built-in support for building remote MCP (Model Context Protocol) servers directly as part of your Agent. This allows you to easily create and manage MCP servers, without the need for additional infrastructure or configuration.

The SDK includes a new MCPAgent class that extends the Agent class and allows you to expose resources and tools over the MCP protocol, as well as authorization and authentication to enable remote MCP servers.

export class MyMCP extends McpAgent {
	server = new McpServer({
		name: "Demo",
		version: "1.0.0",
	});

	async init() {
		this.server.resource(`counter`, `mcp://resource/counter`, (uri) => {
			// ...
		});

		this.server.tool(
			"add",
			"Add two numbers together",
			{ a: z.number(), b: z.number() },
			async ({ a, b }) => {
				// ...
			},
		);
	}
}

export class MyMCP extends McpAgent<Env> {
	server = new McpServer({
		name: "Demo",
		version: "1.0.0",
	});

	async init() {
		this.server.resource(`counter`, `mcp://resource/counter`, (uri) => {
			// ...
		});

		this.server.tool(
			"add",
			"Add two numbers together",
			{ a: z.number(), b: z.number() },
			async ({ a, b }) => {
				// ...
			},
		);
	}
}
``` …

AutoRAG is now in open beta, making it easy for you to build fully-managed retrieval-augmented generation (RAG) pipelines without managing infrastructure. Just upload your docs to R2, and AutoRAG handles the rest: embeddings, indexing, retrieval, and response generation via API.

With AutoRAG, you can:

• Customize your pipeline: Choose from Workers AI models, configure chunking strategies, edit system prompts, and more.
• Instant setup: AutoRAG provisions everything you need from Vectorize, AI gateway, to pipeline logic for you, so you can go from zero to a working RAG pipeline in seconds.
• Keep your index fresh: AutoRAG continuously syncs your index with your data source to ensure responses stay accurate and up to date.
• Ask questions: Query your data and receive grounded responses via a Workers binding or API.

Whether you're building internal tools, AI-powered search, or a support assistant, AutoRAG gets you from idea to deployment in minutes. …

We’re excited to announce Browser Rendering is now available on the Workers Free plan ↗︎, making it even easier to prototype and experiment with web search and headless browser use-cases when building applications on Workers.

The Browser Rendering REST API is now Generally Available, allowing you to control browser instances from outside of Workers applications. We've added three new endpoints to help automate more browser tasks:

• Extract structured data – Use /json to retrieve structured data from a webpage.
• Retrieve links – Use /links to pull all links from a webpage.
• Convert to Markdown – Use /markdown to convert webpage content into Markdown format.

For example, to fetch the Markdown representation of a webpage:

Markdown examplebash

curl -X 'POST' 'https://api.cloudflare.com/client/v4/accounts/<accountId>/browser-rendering/markdown' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer <apiToken>' \
  -d '{
    "url": "https://example.com"
  }'

For the full list of endpoints, check out our REST API documentation. You can also interact with Browser Rendering via the Cloudflare TypeScript SDK ↗︎. …

Durable Objects can now be used with zero commitment on the Workers Free plan allowing you to build AI agents with Agents SDK, collaboration tools, and real-time applications like chat or multiplayer games.

Durable Objects let you build stateful, serverless applications with millions of tiny coordination instances that run your application code alongside (in the same thread!) your durable storage. Each Durable Object can access its own SQLite database through a Storage API. A Durable Object class is defined in a Worker script encapsulating the Durable Object's behavior when accessed from a Worker. To try the code below, click the button:

Deploy to Cloudflare

import { DurableObject } from "cloudflare:workers";

// Durable Object
export class MyDurableObject extends DurableObject {
  ...
	async sayHello(name) {
		return `Hello, ${name}!`;
	}
}

// Worker
export default {
	async fetch(request, env) {
		// Every unique ID refers to an individual instance of the Durable Object class
		const id = env.MY_DURABLE_OBJECT.idFromName("foo"); …

SQLite in Durable Objects is now generally available (GA) with 10GB SQLite database per Durable Object. Since the public beta ↗︎ in September 2024, we've added feature parity and robustness for the SQLite storage backend compared to the preexisting key-value (KV) storage backend for Durable Objects.

SQLite-backed Durable Objects are recommended for all new Durable Object classes, using new_sqlite_classes Wrangler configuration. Only SQLite-backed Durable Objects have access to Storage API's SQL and point-in-time recovery methods, which provide relational data modeling, SQL querying, and better data management.

export class MyDurableObject extends DurableObject {
  sql: SqlStorage
  constructor(ctx: DurableObjectState, env: Env) {
    super(ctx, env);
    this.sql = ctx.storage.sql;
  }

  async sayHello() {
    let result = this.sql
      .exec("SELECT 'Hello, World!' AS greeting")
      .one();
    return result.greeting;
  }
}
``` …

You can now capture a maximum of 256 KB of log events per Workers invocation, helping you gain better visibility into application behavior.

All console.log() statements, exceptions, request metadata, and headers are automatically captured during the Worker invocation and emitted as JSON object. Workers Logs deserializes this object before indexing the fields and storing them. You can also capture, transform, and export the JSON object in a Tail Worker.

256 KB is a 2x increase from the previous 128 KB limit. After you exceed this limit, further context associated with the request will not be recorded in your logs.

This limit is automatically applied to all Workers.

Workflows is now Generally Available (or "GA"): in short, it's ready for production workloads. Alongside marking Workflows as GA, we've introduced a number of changes during the beta period, including:

• A new waitForEvent API that allows a Workflow to wait for an event to occur before continuing execution.
• Increased concurrency: you can run up to 4,500 Workflow instances concurrently — and this will continue to grow.
• Improved observability, including new CPU time metrics that allow you to better understand which Workflow instances are consuming the most resources and/or contributing to your bill.
• Support for vitest for testing Workflows locally and in CI/CD pipelines.

Workflows also supports the new increased CPU limits that apply to Workers, allowing you to run more CPU-intensive tasks (up to 5 minutes of CPU time per instance), not including the time spent waiting on network calls, AI models, or other I/O bound tasks.

Human-in-the-loop

The new step.waitForEvent API allows a Workflow instance to wait on events and data, enabling human-in-the-the-loop interactions, such as approving or rejecting a request, directly handling webhooks from other systems, or pushing event data to a Workflow while it's running. …

We're excited to share that you can now use Playwright's browser automation capabilities ↗︎ from Cloudflare Workers.

Playwright ↗︎ is an open-source package developed by Microsoft that can do browser automation tasks; it's commonly used to write software tests, debug applications, create screenshots, and crawl pages. Like Puppeteer, we forked ↗︎ Playwright and modified it to be compatible with Cloudflare Workers and Browser Rendering.

Below is an example of how to use Playwright with Browser Rendering to test a TODO application using assertions:

Assertion examplets

import { launch, type BrowserWorker } from "@cloudflare/playwright";
import { expect } from "@cloudflare/playwright/test";

interface Env {
	MYBROWSER: BrowserWorker;
}

export default {
	async fetch(request: Request, env: Env) {
		const browser = await launch(env.MYBROWSER);
		const page = await browser.newPage();

		await page.goto("https://demo.playwright.dev/todomvc");

		const TODO_ITEMS = [
			"buy some cheese",
			"feed the cat",
			"book a doctors appointment",
		];

		const newTodo = page.getByPlaceholder("What needs to be done?");
		for (const item of TODO_ITEMS) {
			await newTodo.fill(item); …

Queues now supports the ability to pause message delivery and/or purge (delete) messages on a queue. These operations can be useful when:

• Your consumer has a bug or downtime, and you want to temporarily stop messages from being processed while you fix the bug
• You have pushed invalid messages to a queue due to a code change during development, and you want to clean up the backlog
• Your queue has a backlog that is stale and you want to clean it up to allow new messages to be consumed

To pause a queue using Wrangler, run the pause-delivery command. Paused queues continue to receive messages. And you can easily unpause a queue using the resume-delivery command.

Pause and resume a queuebash

$ wrangler queues pause-delivery my-queue
Pausing message delivery for queue my-queue.
Paused message delivery for queue my-queue.

$ wrangler queues resume-delivery my-queue
Resuming message delivery for queue my-queue.
Resumed message delivery for queue my-queue.

Purging a queue permanently deletes all messages in the queue. Unlike pausing, purging is an irreversible operation:

Purge a queuebash

$ wrangler queues purge my-queue
✔ This operation will permanently delete all the messages in queue my-queue. Type my-queue to proceed. … my-queue
Purged queue 'my-queue'
``` …

You can now run a Worker for up to 5 minutes of CPU time for each request.

Previously, each Workers request ran for a maximum of 30 seconds of CPU time — that is the time that a Worker is actually performing a task (we still allowed unlimited wall-clock time, in case you were waiting on slow resources). This meant that some compute-intensive tasks were impossible to do with a Worker. For instance, you might want to take the cryptographic hash of a large file from R2. If this computation ran for over 30 seconds, the Worker request would have timed out.

By default, Workers are still limited to 30 seconds of CPU time. This protects developers from incurring accidental cost due to buggy code.

By changing the cpu_ms value in your Wrangler configuration, you can opt in to any value up to 300,000 (5 minutes).

{
	// ...rest of your configuration...
	"limits": {
		"cpu_ms": 300000,
	},
	// ...rest of your configuration...
}

[limits]
cpu_ms = 300_000

Note

CPU time is the amount of time the CPU actually spends doing work during a given request. If a Worker's request makes a sub-request and waits for that request to come back before doing additional work, this time spent waiting is not counted towards CPU time.

Worker requests could run for more than 30 seconds of total time prior to this change — only CPU time was limited. …

Source maps are now Generally Available (GA). You can now be uploaded with a maximum gzipped size of 15 MB. Previously, the maximum size limit was 15 MB uncompressed.

Source maps help map between the original source code and the transformed/minified code that gets deployed to production. By uploading your source map, you allow Cloudflare to map the stack trace from exceptions onto the original source code making it easier to debug.

Stack Trace without Source Map remapping

With no source maps uploaded: notice how all the Javascript has been minified to one file, so the stack trace is missing information on file name, shows incorrect line numbers, and incorrectly references js instead of ts.

Stack Trace with Source Map remapping

With source maps uploaded: all methods reference the correct files and line numbers.

Uploading source maps and stack trace remapping happens out of band from the Worker execution, so source maps do not affect upload speed, bundle size, or cold starts. The remapped stack traces are accessible through Tail Workers, Workers Logs, and Workers Logpush. …

Update: Mon Mar 24th, 11PM UTC: Next.js has made further changes to address a smaller vulnerability introduced in the patches made to its middleware handling. Users should upgrade to Next.js versions 15.2.4, 14.2.26, 13.5.10 or 12.3.6. If you are unable to immediately upgrade or are running an older version of Next.js, you can enable the WAF rule described in this changelog as a mitigation.

Update: Mon Mar 24th, 8PM UTC: Next.js has now backported the patch for this vulnerability ↗︎ to cover Next.js v12 and v13. Users on those versions will need to patch to 13.5.9 and 12.3.5 (respectively) to mitigate the vulnerability.

Update: Sat Mar 22nd, 4PM UTC: We have changed this WAF rule to opt-in only, as sites that use auth middleware with third-party auth vendors were observing failing requests.

We strongly recommend updating your version of Next.js (if eligible) to the patched versions, as your app will otherwise be vulnerable to an authentication bypass attack regardless of auth provider.

Enable the Managed Rule (strongly recommended)

This rule is opt-in only for sites on the Pro plan or above in the WAF managed ruleset.

To enable the rule:

1. Head to Security > WAF > Managed rules in the Cloudflare dashboard for the zone (website) you want to protect. …

Smart Placement is a unique Cloudflare feature that can make decisions to move your Worker to run in a more optimal location (such as closer to a database). Instead of always running in the default location (the one closest to where the request is received), Smart Placement uses certain “heuristics” (rules and thresholds) to decide if a different location might be faster or more efficient.

Previously, if these heuristics weren't consistently met, your Worker would revert to running in the default location—even after it had been optimally placed. This meant that if your Worker received minimal traffic for a period of time, the system would reset to the default location, rather than remaining in the optimal one.

Now, once Smart Placement has identified and assigned an optimal location, temporarily dropping below the heuristic thresholds will not force a return to default locations. For example in the previous algorithm, a drop in requests for a few days might return to default locations and heuristics would have to be met again. This was problematic for workloads that made requests to a geographically located resource every few days or longer. In this scenario, your Worker would never get placed optimally. This is no longer the case.

We are excited to announce that AI Gateway now supports real-time AI interactions with the new Realtime WebSockets API.

This new capability allows developers to establish persistent, low-latency connections between their applications and AI models, enabling natural, real-time conversational AI experiences, including speech-to-speech interactions.

The Realtime WebSockets API works with the OpenAI Realtime API ↗︎, Google Gemini Live API ↗︎, and supports real-time text and speech interactions with models from Cartesia ↗︎, and ElevenLabs ↗︎.

Here's how you can connect AI Gateway to OpenAI's Realtime API ↗︎ using WebSockets:

OpenAI Realtime API examplejavascript

import WebSocket from "ws";

const url =
	"wss://gateway.ai.cloudflare.com/v1/<account_id>/<gateway>/openai?model=gpt-4o-realtime-preview-2024-12-17";
const ws = new WebSocket(url, {
	headers: {
		"cf-aig-authorization": process.env.CLOUDFLARE_API_KEY,
		Authorization: "Bearer " + process.env.OPENAI_API_KEY,
		"OpenAI-Beta": "realtime=v1", …

Document conversion plays an important role when designing and developing AI applications and agents. Workers AI now provides the toMarkdown utility method that developers can use to for quick, easy, and convenient conversion and summary of documents in multiple formats to Markdown language.

You can call this new tool using a binding by calling env.AI.toMarkdown() or the using the REST API endpoint.

In this example, we fetch a PDF document and an image from R2 and feed them both to env.AI.toMarkdown(). The result is a list of converted documents. Workers AI models are used automatically to detect and summarize the image.

import { Env } from "./env";

export default {
	async fetch(request: Request, env: Env, ctx: ExecutionContext) {
		// https://pub-979cb28270cc461d94bc8a169d8f389d.r2.dev/somatosensory.pdf
		const pdf = await env.R2.get("somatosensory.pdf");

		// https://pub-979cb28270cc461d94bc8a169d8f389d.r2.dev/cat.jpeg
		const cat = await env.R2.get("cat.jpeg");

		return Response.json(
			await env.AI.toMarkdown([
				{
					name: "somatosensory.pdf",
					blob: new Blob([await pdf.arrayBuffer()], {
						type: "application/octet-stream",
					}),
				},
				{
					name: "cat.jpeg",
					blob: new Blob([await cat.arrayBuffer()], {
						type: "application/octet-stream",
					}),
				},
			]),
		);
	},
};

This is the result:

[
	{
		"name": "somatosensory.pdf",
		"mimeType": "application/pdf", …

npm i agents

agents-sdk -> agents Updated

📝 We've renamed the Agents package to agents!

If you've already been building with the Agents SDK, you can update your dependencies to use the new package name, and replace references to agents-sdk with agents:

# Install the new package
npm i agents

# Remove the old (deprecated) package
npm uninstall agents-sdk

# Find instances of the old package name in your codebase
grep -r 'agents-sdk' .
# Replace instances of the old package name with the new one
# (or use find-replace in your editor)
sed -i 's/agents-sdk/agents/g' $(grep -rl 'agents-sdk' .)

All future updates will be pushed to the new agents package, and the older package has been marked as deprecated.

Agents SDK updates New

We've added a number of big new features to the Agents SDK over the past few weeks, including:

• You can now set cors: true when using routeAgentRequest to return permissive default CORS headers to Agent responses.
• The regular client now syncs state on the agent (just like the React version).
• useAgentChat bug fixes for passing headers/credentials, including properly clearing cache on unmount.
• Experimental /schedule module with a prompt/schema for adding scheduling to your app (with evals!). …

Workers AI is excited to add 4 new models to the catalog, including 2 brand new classes of models with a text-to-speech and reranker model. Introducing:

• @cf/baai/bge-m3 - a multi-lingual embeddings model that supports over 100 languages. It can also simultaneously perform dense retrieval, multi-vector retrieval, and sparse retrieval, with the ability to process inputs of different granularities.
• @cf/baai/bge-reranker-base - our first reranker model! Rerankers are a type of text classification model that takes a query and context, and outputs a similarity score between the two. When used in RAG systems, you can use a reranker after the initial vector search to find the most relevant documents to return to a user by reranking the outputs.
• @cf/openai/whisper-large-v3-turbo - a faster, more accurate speech-to-text model. This model was added earlier but is graduating out of beta with pricing included today.
• @cf/myshell-ai/melotts - our first text-to-speech model that allows users to generate an MP3 with voice audio from inputted text.

Pricing is available for each of these models on the Workers AI pricing page. …

You can now access bindings from anywhere in your Worker by importing the env object from cloudflare:workers.

Previously, env could only be accessed during a request. This meant that bindings could not be used in the top-level context of a Worker.

Now, you can import env and access bindings such as secrets or environment variables in the initial setup for your Worker:

import { env } from "cloudflare:workers";
import ApiClient from "example-api-client";

// API_KEY and LOG_LEVEL now usable in top-level scope
const apiClient = ApiClient.new({ apiKey: env.API_KEY });
const LOG_LEVEL = env.LOG_LEVEL || "info";

export default {
	fetch(req) {
		// you can use apiClient or LOG_LEVEL, configured before any request is handled
	},
};

Note

Workers do not allow I/O from outside a request context. This means that even though env is accessible from the top-level scope, you will not be able to access every binding's methods. …

You can now retry your Cloudflare Pages and Workers builds directly from GitHub. No need to switch to the Cloudflare Dashboard for a simple retry!

Let\u2019s say you push a commit, but your build fails due to a spurious error like a network timeout. Instead of going to the Cloudflare Dashboard to manually retry, you can now rerun the build with just a few clicks inside GitHub, keeping you inside your workflow.

For Pages and Workers projects connected to a GitHub repository:

1. When a build fails, go to your GitHub repository or pull request
2. Select the failed Check Run for the build
3. Select "Details" on the Check Run
4. Select "Rerun" to trigger a retry build for that commit

Learn more about Pages Builds and Workers Builds.

You can now access bindings from anywhere in your Worker by importing the env object from cloudflare:workers.

Previously, env could only be accessed during a request. This meant that bindings could not be used in the top-level context of a Worker.

Now, you can import env and access bindings such as secrets or environment variables in the initial setup for your Worker:

import { env } from "cloudflare:workers";
import ApiClient from "example-api-client";

// API_KEY and LOG_LEVEL now usable in top-level scope
const apiClient = ApiClient.new({ apiKey: env.API_KEY });
const LOG_LEVEL = env.LOG_LEVEL || "info";

export default {
	fetch(req) {
		// you can use apiClient or LOG_LEVEL, configured before any request is handled
	},
};

Note

Workers do not allow I/O from outside a request context. This means that even though env is accessible from the top-level scope, you will not be able to access every binding's methods. …