So ... pretty much none Keep in mind, extensions can update themselves at any time, including when they're bought out by someone else. In fact, I bet that's a huge draw... imagine buying an extension that "can read and modify data on all your websites" and then pushing an update that, oh I dunno, exfiltrates everyone's passwords from their gmail. How would most people even catch that? DO NOT have any extensions runni…
Keep in mind, extensions can update themselves at any time
GP suggested only installing extensions you can build yourself from source. Most extensions that auto update do so via the Chrome store. If you install an extension from source, that won't happen.
What I don't like about LLMs is that people keep re-inventing the wheel over and over. For example, we've been able to control browsers using GPT for about 2 years now: - https://github.com/mayt/BrowserGPT - https://github.com/TaxyAI/browser-extension - https://github.com/browser-use/browser-use - https://github.com/Skyvern-AI/skyvern - https://github.com/m1guelpf/browser-agent - https://github.com/richardyc/Chrome-G…
none of these have stuck right. And none of them work well enough that all web dev agencies no longer have to worry about e2e testing. (or do some of them? Maybe the market is simply that inefficient).
I don't see this being a solution for full e2e regression testing. Having to run inference for each command/test seems expensive. I do think there's room for self-healing tests after failure.
Hey, creator of Browser MCP here. 1. Yes, the extension uses an anonymous device ID and sends an analytics event when a tool call is used. You can inspect the network traffic to verify that zero personalized or identifying information is sent. I collect anonymized usage data to get an idea of how often people are using the extension in the same way that websites count visitors. I split my time between many projects a…
When people see “I collect” they won’t even bother reading further. This is showstopper. Noble reasons won’t matter. Spyware perception.
This seems to be the opposite of what happens in reality.
"detailed" is an anonymized deviceId and a counter of tool calls? Heaven forbid an app want to get some basic insights into how people use it.
This automatic sense of entitlement to surveil users is the absolute embodiment of the banality of evil. It's 2025 - we want informed consent and voluntary participation with the default assumption that no, we do not want you watching over our shoulders, and no, you are not entitled to covertly harvest all the data you want and monetize that without notifying users or asking permissions. The whole ToS gotcha game is…
You're making a huge jump from "gathering anonymous counters to understand how many people use the thing" to "harvest all the data you want and monetize it".
If they were tracking my identity across sites and actually selling it to the highest bidder that's one thing that we'll definitely agree on. This is so so far from that.
You're welcome to build and use your own MCP browser automation if you're so hostile to the developer that built something cool and free for you to use.
"detailed" is an anonymized deviceId and a counter of tool calls? Heaven forbid an app want to get some basic insights into how people use it.
Correct. Telemetry should _always_ be opt-in and explicitly an easy choice to not engage. Any other mode of operation is morally bankrupt.
Really? The hyperbole does not help anyone here.
I don't sign a term sheet when I order at McDonalds but you can be damn sure they count how many big macs I order. Does that make them morally bankrupt? Or is it just a normal business operation that is actually totally reasonable?
none of these have stuck right. And none of them work well enough that all web dev agencies no longer have to worry about e2e testing. (or do some of them? Maybe the market is simply that inefficient).
I don't see this being a solution for full e2e regression testing. Having to run inference for each command/test seems expensive. I do think there's room for self-healing tests after failure.
If this works well enough couldn't you save the selectors and only use inference running the test for the first time and when the UI has changed. Cheaper than dev?
Crazy, in looking up some info on the web and creating a Spreadsheet on Google Sheets to insert the results, it worked almost perfectly the first time and completely failed subsequently on 8-10 different tries. Is there an issue with the lag between what is happening in the browser and the MCP app (in my case Claude Desktop)? I have a feeling the first time I tried it, I was fast enough clicking the "Allow for this c…
What you're experiencing is commonly referred to as "luck". It's the same reason people consistently think newer versions of ChatGPT are nerfed in some way. In reality, people just got lucky originally and have unrealistic expectations based on this originally positive outcome. There's no bug or glitch happening. It's just statistically unlikely to perform the action you wanted and you landed a good dice roll on your…
haha yeh as someone who has built automation for years i can agree with this. You cant just click on something in a script, you need to reliably click on something. As a user, its very easy for you to make adjustments like clicking twice on a link if it doesnt load in time. Thats pretty much what your automation suite needs to end up with. A series of a functions to emulate user actions. You then combine that together with your scripts to create reliable scripts that can run in different conditions. LLMs wont do that for you, u need to instruct them specifically.
Correct. Telemetry should _always_ be opt-in and explicitly an easy choice to not engage. Any other mode of operation is morally bankrupt.
Really? The hyperbole does not help anyone here. I don't sign a term sheet when I order at McDonalds but you can be damn sure they count how many big macs I order. Does that make them morally bankrupt? Or is it just a normal business operation that is actually totally reasonable?
> The moment tools like socket2 or HTTP clients land with Preview2 i'm waiting for that as well. my other options are - either bind a host function to manage wss connection to wasm. fork a CDP lib to use that. - create a proxy between http/wss maybe. And then fork a CDP lib to use http proxy i think.