Can someone explain what the hell is going on here?
Do websites want to prevent automated tooling, as indicated by everyone putting everything behind Cloudfare and CAPTCHAs since forever, or do websites want you to be able to automate things? Because I don't see how you can have both.
If I'm using Selenium it's a problem, but if I'm using Claude it's fine??
In a nutshell: Google wants your websites to be more easily used by the agents they are putting in the browser and other products.
They own the user layer and models, and get to decide if your product will be used.
Think search monopoly, except your site doesn't even exist as far as users are concerned, it's only used via an agent, and only if Google allows.
The work of implementing this is on you. Google is building the hooks into the browser for you to do it; that's WebMCP.
It's all opaque; any oopsies/dark patterns will be blamed on the AI. The profits (and future ad revenue charged for sites to show up on the LLM's radar) will be claimed by Google.
The other AI companies are on board with this plan. Any questions?
The irony is Google properties are more locked down than ever. When I use a commercial VPN I get ReCAPTCHA’ed half of the time doing every single Google search; and can’t use YouTube in Incognito sometimes, “Sign in to confirm you’re not a bot”.
We should definitely feel trepidation at the prospects of any LLM guided browser, in addition to WebMCP (e.g. Claude for Chrome enters the same opaque LLM-controlled/deferred decision process, OpenClaw etc).
Just one example: Prompting the browser to "register example.com" means that Google/Anthropic gets to hustle registrars for SEO-style priority. Using countermeasures like captcha locks you out of the LLM market.
Google's incentive to allow you to shop around via traditional web search is decreased since traditional ads won't be as lucrative (businesses will catch on that blanket targeted ads aren't as effective as a "referral" that directs an LLM to sign-up/purchase/exchange something directly)... expect web search quality to decline, perhaps intentionally.
The only way to combat this, as far as I can conceptualize, is with open models, which are not yet as good as private ones, in no small part due to the extraordinary investment subsidization. We can hope for the bubble to pop, but plan for a deader Internet.
Meanwhile, trust online, at large, begins to evaporate as nobody can tell what is an LLM vs a human-conducted browser. The Internet at large is entering some very dark waters.
> Do websites want to prevent automated tooling, as indicated by everyone putting everything behind Cloudfare and CAPTCHAs since forever, or do websites want you to be able to automate things? Because I don't see how you can have both.
> Since tool calls are handled in JavaScript, a browsing context (i.e. a browser tab or a webview) must be opened. There is no support for agents or assistive tools to call tools "headlessly," meaning without visible browser UI.
I'm old enough to remember discussions around the meaning of `User-Agent` and why it was important that we include it in HTTP headers. Back before it was locked to `Chromium (Gecko; Mozilla 4.0/NetScape; 147.01 ...)`. We talked about a magical future where your PDA, car, or autonomous toaster could be browsing the web on your behalf, and consuming (or not consuming) the delivered HTML as necessary. Back when we named it "user agent" on purpose. AI tooling can finally realize this for the Web, but it's a shame that so many companies who built their empires on the shoulders of those visionaries think the only valid way to browse is with a human-eyeball-to-server chain of trust.
Me too but it died when ads became the currency of the web. If the reason the site exists is to use ads, they’re not going to let you use an user agent that doesn’t display the ads.
> If the reason the site exists is to use ads, they’re not going to let you use an user agent that doesn’t display the ads.
They've been giving it the old college try for the better part of two decades and the only website I've had to train myself not to visit is Twitch, whose ads have invaded my sightline one time too many, and I conceded that particular adblocking battle. I don't get the sense that it's high on the priority list for most sites out there (knock on wood).
Just like then we were naive about folks not abusing these things to the point of making everyone need to block them to oblivion. I think we are relearning these lessons 30 years later.
i’m seeing this at my corporate software job now. that service that you used to have security and product approval for to even read their Swagger doc has an MCP server you can install with 2 clicks.
It’s weirder than that. There is a surge of companies working on how to provide automated access to things like payments, email, signup flows, etc to *Claw.
Also, as someone who has tried to build tools that automate finding flights, The existing players in the space have made it nearly impossible to do. But now Google is just going to open the door for it?
I feel like this is a way to ultimately limit the ability to scrape but also the ability to use your own AI agent to take actions across the internet for you. Like how Amazon doesn’t let your agent to shop their site for you, but they’ll happily scrape every competitor’s website to enforce their anti competitive price fixing scheme. They want to allow and deny access on their terms.
WebMCP will become another channel controlled by big tech and it’ll come with controls. First they’ll lure people to use this method for the situations they want to allow, and then they’ll block everything else.
Oh, that's an easy one. LLMs have made people lose their god damned minds. It makes sense when you think about it as breaking a few eggs to get to the promised land omelette of laying off the development staff.
In my opinion sites that want agent access should expose server-side MCP, server owns the tools, no browser middleman. Already works today.
Sites that don’t want it will keep blocking. WebMCP doesn’t change that.
Your point about selenium is absolutely right. WebMCP is an unnecessary standard. Same developer effort as server-side MCP but routed through the browser, creating a copy that drifts from the actual UI. For the long tail that won’t build any agent interface, the browser should just get smarter at reading what’s already there.
Hey, it's the semantic web, but with ~~XML~~, ~~AJAX~~, ~~Blockchain~~, Ai!
Well, it has precisely the problem of the semantic web, it asks the website to declare in a machine readable format what the website does. Now, llms are kinda the tool to interface to everybody using a somewhat different standard, and this doesn't need everybody to hop on the bandwagon, so perhaps this is the time where it is different.
I wonder what limitations Google is planning with this API to avoid misuse[0] (from the agent/Google's perspective).
A website that doesn't want to be interfaced by an agent (because they want a human to see their ads) could register bogus but plausible tools that convince the agent that the tool did something good. Perhaps the website could also try prompt injecting the agent into advertising to the user on the website's behalf.
I suspect people will get pretty riled up in the comments. This is fine folks. More people will make their stuff machine-accessible and that's a good thing even if MCP won't last or if it's like VHS -- yes Betamax was better, but VHS pushed home video.
Majority of sites don't even expose accessibility functionalities, and for WebMCP you have to expose and maintain internal APIs per page. This opens the site up to abuse/scraping/etc.
Thats why I dont see this standard going to takeoff.
Google put it out there to see uptake. Its really fun to talk about but will be forgotten by end of year is my hot take.
Rather what I think will be the future is that each website will have its own web agent to conversationally get tasks done on the site without you having to figure out how the site works. This is the thesis for Rover (rover.rtrvr.ai), our embeddable web agent with which any site can add a web agent that can type/click/fill by just adding a script tag.
> for WebMCP you have to expose and maintain internal APIs per page
Perhaps. I think an API for the session is probably the root concern. Page specific is nice to have.
You say it like it's a bad thing. But ideally this also brings clarity & purpose to your own API design too! Ideally there is conjunct purpose! And perhaps shared mechanism!
> This opens the site up to abuse/scraping/etc.
In general it bothers me that this is regarded as a problem at all. In principle, sites that try to clickjack & prevent people from downloading images or whatever have been with us for decades. Trying to keep users from seeing what data they want is, generally, not something I favor.
I'd like to see some positive reward cycles begin, where sites let users do more, enable them to get what they want more quickly, in ways that work better for them.
The web is so unique in that users often can reject being corralled and cajoled. That they have some choice. A lot of businesses being the old app-centric "we determine the user experience" ego to the web when they work, but, imo, there's such a symbiosis to be won by both parties by actually enhancing user agency, rather than this war against your most engaged users.
This also could be a great way to avoid scraping and abuse, by offering a better system of access so people don't feel like they need to scrape your site to get what they want.
> Rather what I think will be the future is that each website will have its own web agent to conversationally get tasks done on the site without you having to figure out how the site works
For someone who just was talking about abuse, this seems like a surprising idea. Your site running its own agent is going to take a lot of resources!! Insuring those resources go to what is mutually beneficial to you both seems... difficult.
It also, imo, misses the idea of what MCP is. MCP is a tool calling system, and usually, it's not just one tool involved! If an agent is using webmcp to send contacts from one MCP system into a party planning webmcp, that whole flow is interesting and compelling because the agent can orchestrate across multiple systems.
Trying to build your own agent is, broadly, imo, a terrible idea, that will never allow the user to wield the connected agency they would want to be bringing. What's so exciting an interesting about the agent age is that the walls and borders of software are crumbling down, and software is intertwingularizing, is soft & malleable again. You need to meet users & agents where they are at, if you want to participate in this new age of software.
> You say it like it's a bad thing. But ideally this also brings clarity & purpose to your own API design too! Ideally there is conjunct purpose! And perhaps shared mechanism!
I update my website multiple times a day. I want to have as much decoupling as possible. Everytime I update internal API, I dont want to think of having to also update this WebMCP config.
Basically I have to put in work setting up WebMCP, so that Google can have a better agent that disintermediates my site.
> Trying to keep users from seeing what data they want is, generally, not something I favor.
This is literally the whole cat and mouse game of scraping and web automation, sites clearly want to protect their moat and differentiators. LinkedIn/X/Google literally sue people for scraping, I don't think they themselves are going to package all this data as a WebMCP endpoint for easy scraping.
Regardless of your preferences/ideals, the ecosystem is not going to change overnight due to hype about agents.
> Your site running its own agent is going to take a lot of resources
A lot of sites already expose chatbots, its trivial to rate limit and captcha on abuse detection
Sadly I do see this slop taking off purely because something something AI, investors, shareholders, hype. I mean even the Chrome devtools now push AI in my face at least once a week, so the slop has saturated all the layers.
They don't give a fuck about accessibility unless it results in fines. Otherwise it's totally invisible to them. AI on the other hand is everywhere at the moment.
Trust me bro this API is just temporary, soon™ they'll be able to do everything without help... I just need you to implement this one little API for now so NON-VISIONARY people can get a peek at what it'll look like in 3 months. PLEASE BRO.
I actually think webmcp is incredibly smart & good (giving users agency over what's happening on the page is a giant leap forward for users vs exposing APIs).
But this post frustrates the hell out of me. There's no code! An incredibly brief barely technical run-down of declarative vs imperative is the bulk of the "technical" content. No follow up links even!
I find this developer.chrome.com post to be broadly insulting. It has no on-ramps for developers.
Between Zero Click Internet (AI Summaries) + WebMCP (Dead Internet) why should content producers produce anything that's not behind a paywall the days?
Well I have had the problem of "I want to find the cheapest flight that leaves during this range of dates, and returns during this range of dates, but isn't early in the morning or late at night, and includes additional fees for the luggage I need in the price comparison" and current search tools can't do that very well. I'm not very optimistic WebMCP would solve that though.
I dind't know about it, just checked it out for a flight I'll buy soon, and has almost no direct flights which I know exist because they're on skyscanner...
I want my local dm shop to offer me their product info as copyable markdown, ingredient list, and other health related information. This could be a way to automate it.
Given that it's about food or medicine somehow, because of the mention of ingredients lists and health-related information, it's probably https://en.wikipedia.org/wiki/Dm-drogerie_markt (usually abbreviated "dm").
Do websites want to prevent automated tooling, as indicated by everyone putting everything behind Cloudfare and CAPTCHAs since forever, or do websites want you to be able to automate things? Because I don't see how you can have both.
If I'm using Selenium it's a problem, but if I'm using Claude it's fine??
They own the user layer and models, and get to decide if your product will be used.
Think search monopoly, except your site doesn't even exist as far as users are concerned, it's only used via an agent, and only if Google allows.
The work of implementing this is on you. Google is building the hooks into the browser for you to do it; that's WebMCP.
It's all opaque; any oopsies/dark patterns will be blamed on the AI. The profits (and future ad revenue charged for sites to show up on the LLM's radar) will be claimed by Google.
The other AI companies are on board with this plan. Any questions?
Just one example: Prompting the browser to "register example.com" means that Google/Anthropic gets to hustle registrars for SEO-style priority. Using countermeasures like captcha locks you out of the LLM market.
Google's incentive to allow you to shop around via traditional web search is decreased since traditional ads won't be as lucrative (businesses will catch on that blanket targeted ads aren't as effective as a "referral" that directs an LLM to sign-up/purchase/exchange something directly)... expect web search quality to decline, perhaps intentionally.
The only way to combat this, as far as I can conceptualize, is with open models, which are not yet as good as private ones, in no small part due to the extraordinary investment subsidization. We can hope for the bubble to pop, but plan for a deader Internet.
Meanwhile, trust online, at large, begins to evaporate as nobody can tell what is an LLM vs a human-conducted browser. The Internet at large is entering some very dark waters.
The docs (https://docs.google.com/document/d/1rtU1fRPS0bMqd9abMG_hc6K9...) draw the line at headless automation: it requires a visible browsing context:
> Since tool calls are handled in JavaScript, a browsing context (i.e. a browser tab or a webview) must be opened. There is no support for agents or assistive tools to call tools "headlessly," meaning without visible browser UI.
They've been giving it the old college try for the better part of two decades and the only website I've had to train myself not to visit is Twitch, whose ads have invaded my sightline one time too many, and I conceded that particular adblocking battle. I don't get the sense that it's high on the priority list for most sites out there (knock on wood).
An e-commerce? Wanna automate buying your stuff - probably something they wanna allow under controlled forms
Wanna scrape the site to compare prices? Maybe less so.
Also I just recently noticed Chrome now has a Klarna/BNPL thing as a built in payments option that I never asked for...
WebMCP will become another channel controlled by big tech and it’ll come with controls. First they’ll lure people to use this method for the situations they want to allow, and then they’ll block everything else.
Sites that don’t want it will keep blocking. WebMCP doesn’t change that.
Your point about selenium is absolutely right. WebMCP is an unnecessary standard. Same developer effort as server-side MCP but routed through the browser, creating a copy that drifts from the actual UI. For the long tail that won’t build any agent interface, the browser should just get smarter at reading what’s already there.
Wrote about it here: https://open.substack.com/pub/manveerc/p/webmcp-false-econom...
Most sites don't want to expose APIs or care enough about setup and maintenance of said API.
Well, it has precisely the problem of the semantic web, it asks the website to declare in a machine readable format what the website does. Now, llms are kinda the tool to interface to everybody using a somewhat different standard, and this doesn't need everybody to hop on the bandwagon, so perhaps this is the time where it is different.
A website that doesn't want to be interfaced by an agent (because they want a human to see their ads) could register bogus but plausible tools that convince the agent that the tool did something good. Perhaps the website could also try prompt injecting the agent into advertising to the user on the website's behalf.
[0]: Beyond just hoping the website complies with their "Generative AI Prohibited Uses Policy": https://developer.chrome.com/docs/ai/get-started#gemini_nano...
Thats why I dont see this standard going to takeoff.
Google put it out there to see uptake. Its really fun to talk about but will be forgotten by end of year is my hot take.
Rather what I think will be the future is that each website will have its own web agent to conversationally get tasks done on the site without you having to figure out how the site works. This is the thesis for Rover (rover.rtrvr.ai), our embeddable web agent with which any site can add a web agent that can type/click/fill by just adding a script tag.
IYKYK
Perhaps. I think an API for the session is probably the root concern. Page specific is nice to have.
You say it like it's a bad thing. But ideally this also brings clarity & purpose to your own API design too! Ideally there is conjunct purpose! And perhaps shared mechanism!
> This opens the site up to abuse/scraping/etc.
In general it bothers me that this is regarded as a problem at all. In principle, sites that try to clickjack & prevent people from downloading images or whatever have been with us for decades. Trying to keep users from seeing what data they want is, generally, not something I favor.
I'd like to see some positive reward cycles begin, where sites let users do more, enable them to get what they want more quickly, in ways that work better for them.
The web is so unique in that users often can reject being corralled and cajoled. That they have some choice. A lot of businesses being the old app-centric "we determine the user experience" ego to the web when they work, but, imo, there's such a symbiosis to be won by both parties by actually enhancing user agency, rather than this war against your most engaged users.
This also could be a great way to avoid scraping and abuse, by offering a better system of access so people don't feel like they need to scrape your site to get what they want.
> Rather what I think will be the future is that each website will have its own web agent to conversationally get tasks done on the site without you having to figure out how the site works
For someone who just was talking about abuse, this seems like a surprising idea. Your site running its own agent is going to take a lot of resources!! Insuring those resources go to what is mutually beneficial to you both seems... difficult.
It also, imo, misses the idea of what MCP is. MCP is a tool calling system, and usually, it's not just one tool involved! If an agent is using webmcp to send contacts from one MCP system into a party planning webmcp, that whole flow is interesting and compelling because the agent can orchestrate across multiple systems.
Trying to build your own agent is, broadly, imo, a terrible idea, that will never allow the user to wield the connected agency they would want to be bringing. What's so exciting an interesting about the agent age is that the walls and borders of software are crumbling down, and software is intertwingularizing, is soft & malleable again. You need to meet users & agents where they are at, if you want to participate in this new age of software.
I update my website multiple times a day. I want to have as much decoupling as possible. Everytime I update internal API, I dont want to think of having to also update this WebMCP config.
Basically I have to put in work setting up WebMCP, so that Google can have a better agent that disintermediates my site.
> Trying to keep users from seeing what data they want is, generally, not something I favor.
This is literally the whole cat and mouse game of scraping and web automation, sites clearly want to protect their moat and differentiators. LinkedIn/X/Google literally sue people for scraping, I don't think they themselves are going to package all this data as a WebMCP endpoint for easy scraping.
Regardless of your preferences/ideals, the ecosystem is not going to change overnight due to hype about agents.
> Your site running its own agent is going to take a lot of resources
A lot of sites already expose chatbots, its trivial to rate limit and captcha on abuse detection
They don't give a fuck about accessibility unless it results in fines. Otherwise it's totally invisible to them. AI on the other hand is everywhere at the moment.
But this post frustrates the hell out of me. There's no code! An incredibly brief barely technical run-down of declarative vs imperative is the bulk of the "technical" content. No follow up links even!
I find this developer.chrome.com post to be broadly insulting. It has no on-ramps for developers.
Can we stop pretending this is an issue anyone has ever had.
I dind't know about it, just checked it out for a flight I'll buy soon, and has almost no direct flights which I know exist because they're on skyscanner...
Or maybe it's a "direct marketing shop", where you bring flyers to be delivered into people's mail? Yeah, that must be it.
(I didn't know about that either before now.)
https://www.dm.de/