keep running into rate limits for browser rendering despite paid plan?
keep running into rate limits for browser rendering despite paid plan?
HI'm using it for og image generation
Kif you're running into limits while on paid plan you can request a limit increase here. https://forms.gle/CdueDKvb26mTaepa9
we'll evaluate on a case by case basis
we'll evaluate on a case by case basis
Google Docs
Want to request an increase to a limit from Browser Rendering? Let us know here.
If the limit can be increased, Cloudflare will contact you with next steps.
If the limit can be increased, Cloudflare will contact you with next steps.
Hfilled 
Hthanks
SHey, I am trying to access browser rendering rest api.
client.browserRendering.screenshot.create()
This does not work
While simple fetch without Cloudflare client works.
client.browserRendering.screenshot.create()
This does not work
While simple fetch without Cloudflare client works.
Dhi team (@Rui Figueira ?) i notice https://github.com/cloudflare/playwright hasn't been released for a few months now, but lots of stuff has made it in. what's the plan around this?
i'm also seeing playwright-mcp is pointing to 1.53.0-alpha-1745357020000 and not the cloudflare fork.. how does that work?
i'm also seeing playwright-mcp is pointing to 1.53.0-alpha-1745357020000 and not the cloudflare fork.. how does that work?
GitHub
Playwright fork that works with Cloudflare Browser Rendering - cloudflare/playwright
RHey @dandoen ! We're working on upgrading playwright to a latest version. Actually, our main branch is already in sync with playwright 1.53.0, but are still adding a few more things before releasing
Rregarding playwright-mcp, and to avoid significant changes in the upstream code, we keep a cloudflare folder where all workers logic is, and it also imports our @cloudflare/playwright dependency
GitHub
Playwright MCP fork that works with Cloudflare Browser Rendering - cloudflare/playwright-mcp
Dfantastic, i'll give it a shot to use cloudflare/playwright@main instead as i need some latest features (like the private method _snapshotForAI)
Rnice! I'll try to accelerate those upgrades
Dyou rock, thank you. i'll keep you posted how it goes!
Dactually would it be easy to get a -next release or so? happy to be a guinea pig
Kwhat's your use case? and what latest features are you most interested in+why?
D@Kathy running my own browser DO from which i have my ai agent make tool calls, so can’t use playwright-mcp (violates the DOs i/o rules). hence need to do a few things similar to what playwright mcp dus just in my own service logic (like leveraging private apis from newer versions of playwright to get the accessibility tree for AI). in addition i couldn’t get a few cdp things to work on the 0.11 version, so hoping upgrade would help
hope that helps!
hope that helps!
JAnother ping for this one @Kathy, thanks
Dhello team, when using the cloudflare's playwright mcp from another worker i run into this:
Error: page.goto: Cannot perform I/O on behalf of a different Durable Object. I/O objects (such as streams, request/response bodies, and others) created in the context of one Durable Object cannot be accessed from a different Durable Object in the same isolate. This is a limitation of Cloudflare Workers which allows us to improve overall performance. (I/O type: Native)
note the original request to the mcp isn't made from a DO, so i'm not sure where this is coming from and what i can do to mitigate.
Error: page.goto: Cannot perform I/O on behalf of a different Durable Object. I/O objects (such as streams, request/response bodies, and others) created in the context of one Durable Object cannot be accessed from a different Durable Object in the same isolate. This is a limitation of Cloudflare Workers which allows us to improve overall performance. (I/O type: Native)
note the original request to the mcp isn't made from a DO, so i'm not sure where this is coming from and what i can do to mitigate.
Khad been in progress! https://github.com/cloudflare/puppeteer/releases/tag/v1.0.4
- now done thanks to @Rui Figueira
GitHub
What's Changed
Upgrade to puppeteer v22.13.1 in #132
Full Changelog: https://github.com/cloudflare/puppeteer/compare/v1.0.3..v1.0.4
Upgrade to puppeteer v22.13.1 in #132
Full Changelog: https://github.com/cloudflare/puppeteer/compare/v1.0.3..v1.0.4
Jthanks! :MeowHeartCloudflare:
D@Kathy qq is the playwright-mcp not designed to be used in combination with other cloudflare workers or am i missing something? works greatly when using it as an mcp w/ claude desktop, cursor etc but seems odd to not be able to use it with the agent sdk/vercel's ai sdk running inside cloudflare!
Rcan you provide a code snippet for me to test it?
Dhttps://ai-sdk.dev/docs/reference/ai-sdk-core/create-mcp-client
deploy this snippet from the example to a worker. just instead of type stdio, set type to sse and provide url = other worker url that serves the playwright mcp
deploy this snippet from the example to a worker. just instead of type stdio, set type to sse and provide url = other worker url that serves the playwright mcp
Rhmm, I see... but maybe you can do that without creating the MCP server, because at the end you just need the tools, right?
Di guess not ideal (i’d love to separate and in my app users can bring their own mcp servers so it may be hosted on CF) .but i can give it a try.. any guidance on how to use the mcp lib without creating the server?
Kheads up - content endpoint was updated. and now returns title https://developers.cloudflare.com/api/resources/browser_rendering/subresources/content/methods/create/#(params)%200%20%3E%20(param)%20rejectRequestPattern%20%3E%20(schema)
Interact with Cloudflare's products and services via the Cloudflare API
Dcorrect me if i'm wrong but i only see
createMcpAgent exposing the serve and mount methods. no way to actually use just the tools?Ryes, right now without changes to the library it's not trivial, but all tools are available in playwright-mcp internal code
Dok, thx. i'm going w/ browserbase for now,
R@dandoen btw, just to better understand yhe error you had: are you exposing playwright MCP in the same worker that tries to consume it? Or do you have two different workers / apps, one as a playwright mcp server, and the other as a mcp client using ai-sdk?
Dtwo different workers!
Rwell, I would have to check that better, but nevertheless as I was saying , and because playwright-mcp basically has its toolset locally, there's no need to perform external calls. I did a small experiment of exposing a toolsetProvider, you can take a look at https://github.com/cloudflare/playwright-mcp/pull/20, check the example for usage (this depends on our merge of playwright-mcp 0.0.30 which was not released yet, we only released a new release of playwright today)
Dthat sounds great, i'll check it out.
Dbtw do you ever see cloudflare/playwright not force me to use the browser rendering api? i'd love to be able to use playwright-core inside workers to communicate with either my own CDP endpoint over ws or providers like browser base.
i noticed (at least the currently published 0.10 release) is doing a few things like overriding the host (setting a fake host), expecting session id query parameters to decide when to use connect (easy to bypass), checking for binding in the query param (easy to bypass) but also chunking the ws messages presumably due to limitation in the browser render api.
i noticed (at least the currently published 0.10 release) is doing a few things like overriding the host (setting a fake host), expecting session id query parameters to decide when to use connect (easy to bypass), checking for binding in the query param (easy to bypass) but also chunking the ws messages presumably due to limitation in the browser render api.
Di just spent the day patching playwright-core, in a similar way i saw ya'll had done it to make it work inside the worker runtime 
Rfor the time being it's not possible. We could in theory allow both scenarios, but from my experience, if the browser is not in the same network, we get lots of latency. For instance, a simple 
newPage call triggers 50 CDP calls, and if each takes 50ms, it's a full second just to open a new page Dtrue, but also depends on the customer use case whether 1s is acceptable or not. ie if it's an async job running in the background, then 1s doesn't really matter
DI'm also interested in this!
@Rui Figueira Would you be interested in supporting this?
https://github.com/cloudflare/playwright/pull/59
cc @dandoen
@Rui Figueira Would you be interested in supporting this?
https://github.com/cloudflare/playwright/pull/59
cc @dandoen
GitHub
This adds support for using any ws CDP based server to connect to, without breaking existing usage of the browser rendering api. Example usage with Browserbase:
import { Browserbase } from &#39...
import { Browserbase } from &#39...
D@Kathy curious if ya'll would be interested to make the @Cloudflare/playwright package work also for customers not using the browser rendering API (yet). by supporting this, you'll help me keep an easy migration path (back) to cloudflare
AHi, I found that using the /content endpoint we can not set the returned HTML type. Others like Crawl4ai allow us to set raw, cleaned HTML and markdown. I think CF should implement it.
HIs browser rendering extremely slow for anyone else to bootup/work?
Hseeing extremely slow times.
Kyou're looking for info to be returned in markdown? we have https://developers.cloudflare.com/browser-rendering/rest-api/markdown-endpoint/
Cloudflare Docs
The /markdown endpoint retrieves a webpage's content and converts it into Markdown format. You can specify a URL and optional parameters to refine the extraction process.
AThank you, but I need the other 2 ones: raw HTML for getting invisible metadata and cleaned HTML for visible content. i think both Firecrawl and Crawl4ai features would inspire Browser Rendering PMs and devs a lot.
Pis Browser Rendering down? Been getting 503 error for the last 10 minutes
Kso our /html endpoint returns raw already today. So what we don't yet have for you is
can you explain what you'd like to do with
cleaned html then.can you explain what you'd like to do with
cleaned that raw
Hwe ended up moving away
Hused both playwright/puppeeter, don't think it makes any difference.
SCan we have wrapped functions for rest api in worker bindings?
K
is on the list