[HN Gopher] WebMCP
___________________________________________________________________
WebMCP
Author : sanj
Score : 123 points
Date : 2025-10-17 21:57 UTC (1 days ago)
(HTM) web link (github.com)
(TXT) w3m dump (github.com)
| westurner wrote:
| W3C specs are written with respec: ReSpec docs:
| https://respec.org/docs/#w3c-documents
|
| W3C Process document > 3.4. Chartered Groups: Working Groups and
| Interest Groups: https://www.w3.org/policies/process/#GAGeneral
|
| There's WebGPU, WebNN, window.ai, Prompt API, Summarizer API,
| Writer API, Rewriter API, Language Detector API, Translator API ;
| and now WebMCP
|
| WebNN: https://www.w3.org/TR/webnn/
|
| webmachinelearning/prompt-api > "Explainer for the Prompt API":
| https://github.com/webmachinelearning/prompt-api
|
| https://developer.chrome.com/docs/ai/built-in :
|
| > Standardization effort: _We 're working to standardize all of
| these APIs for cross-browser compatibility._
|
| > _The Language Detector API and Translator API have been adopted
| by the W3C WebML Working Group. We 've asked Mozilla and WebKit
| for their standards positions._
|
| > _The Summarizer API, Writer API, and Rewriter API have also
| been adopted by the W3C WebML Working Group. We 've asked asked
| Mozilla and WebKit for their standards positions._
|
| webmachinelearning/webmcp:
| https://github.com/webmachinelearning/webmcp
|
| jasonjmcghee/WebMCP: https://github.com/jasonjmcghee/WebMCP
|
| Having worked on at least one web app with a name that started
| with "Web", I'm not surprised.
|
| /? mcp chrome:
| https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu... :
|
| - "Show HN: We packaged an MCP server inside Chromium" (today)
| https://news.ycombinator.com/item?id=45618536 re: browseros-mcp:
| https://github.com/browseros-ai/BrowserOS/blob/main/docs/bro...
|
| - "Chrome DevTools (MCP) for your AI agent"
| https://developer.chrome.com/blog/chrome-devtools-mcp ..
| https://news.ycombinator.com/item?id=45412734 (September 2025) ..
| :
|
| > _We 're launching today a public preview for the new Chrome
| DevTools Model Context Protocol (MCP) server, bringing the power
| of Chrome DevTools to AI coding assistants._
|
| > _Coding agents face a fundamental problem: they are not able to
| see what the code they generate actually does when it runs in the
| browser. They 're effectively programming with a blindfold on._
|
| > _The Chrome DevTools MCP server changes this. AI coding
| assistants are able to debug web pages directly in Chrome, and
| benefit from DevTools debugging capabilities and performance
| insights. This improves their accuracy when identifying and
| fixing issues._
|
| How could the Chrome DevTools MCP be integrated with the Gemini
| Computer Use model?
|
| From https://news.ycombinator.com/item?id=45543923 :
|
| > Competency Story: _The customer and product owner can write BDD
| tests in order to validate the app against the requirements_
|
| > Prompt: _Write playwright tests for #token_reference, that run
| a named factored-out login sequence, and then test as human user
| would that: when you click on Home that it navigates to / (given
| browser MCP and recently the_ Gemini 2.5 Computer Operator model
| _)_
|
| "Introducing the Gemini 2.5 Computer Use model" (October 2025)
| https://blog.google/technology/google-deepmind/gemini-comput...
|
| Could this help with accessibility reviews?
|
| "Lighthouse accessibility score"
| https://developer.chrome.com/docs/lighthouse/accessibility/s...
|
| awesome-a11y > Tools:
| https://github.com/brunopulis/awesome-a11y/blob/main/topics/...
| ocdtrekkie wrote:
| This explains all the new random GPO settings I had to go
| disable at the office this week! (A lot of users are reporting
| performance issues with browsers, seems like all the browsers
| are adding AI things... seems like a good place to start.)
| westurner wrote:
| This is as bad or worse than agreeing to voice search with
| them.
|
| Hadn't realized we've all been opted-in.
|
| My voice assistant used to be able to create a reminder
| without siphoning everything out to "must be reviewed because
| it's AI" remote AI.
|
| Is it possible to use non-AI voice search on YouTube (with
| GoogleTV) without signing one's life away?
|
| Try voice searching for "weather in [city]" with YT on GTV:
| it launches another (Google) app instead of just adding text
| to the search field.
|
| When they asked for suggestions for OpenAI's fork of
| Chromium, I suggested adding fuzzy and regex search in a
| drawer and sending it upstream; like vimgrep for Chromium.
| That would help solve for _Search_ , like the original
| mission of the company.
| jasonjmcghee wrote:
| Surprised to see this on HN front-page.
|
| A lot has happened since I proposed / built this.
|
| WebMCP is being incubated in W3C / webmachinelearning, so highly
| recommend checking that out as it's what will turn into WebMCP
| being in your browser.
|
| https://github.com/webmachinelearning/webmcp
| koolala wrote:
| How much of this can or can't be done by a Extension or
| Bookmarklet running JS in a page? What is your biggest dream
| from standardizing this?
| nsonha wrote:
| it seems like a cleaner approach to declare a handful of
| tools that users can approve/ask for granularily, than just
| say "my website can run any wacky script, here is some
| bookmarklet, nerds" or the very generic permission model of
| browser extensions
| koolala wrote:
| Couldn't that be solved with a simple bookmarklet
| permission model? The script would request the same pop-up
| features as a website?
| miguelspizza wrote:
| It's more providing permission granularity on the action
| level rather than the sandbox level. Your script might
| not be able to make external api calls, but there is no
| way to gate the ability to take destructive action within
| the webpage.
|
| With something like WebMCP you get elicitation and the
| ability to disable tools from the client.
| koolala wrote:
| What kind of destructive action do you mean that is so
| critical?
| miguelspizza wrote:
| WebMCP essentially turns your website into an MCP server.
| Which means it is kind of like building a UI for the LLM
| that lives alongside the human UI.
|
| It's also a contract for how LLM's interact with a
| website, they can do no more than the tools allow them to
| do. When you are running javascript on the page, the
| entire website is an attack surface.
|
| Let's take gmail, for example. There is no way to protect
| your webpage from an agent running a script that sends an
| email by triggering the send email button. But with
| WebMCP, you can explicitly disable the "send_email" tool
| when the agent interacts with gmail.
| koolala wrote:
| That sounds nice. That makes the WebMCP both capabilities
| and permissions.
| jasonjmcghee wrote:
| If there's a standard instead of needing to download the Jira
| mcp server, you just visit their website and all the tools
| are described and usable from there.
|
| Or put differently, as a company / group / individual,
| instead of needing to build and distribute an mcp server and
| public API, you can just support WebMCP.
|
| Another alternative is LLMs / agents operating playwright or
| the equivalent which will likely be less reliable and consume
| more tokens. (By a fair margin)
| tobyjsullivan wrote:
| I had to skim through this a couple times before I realized that
| I still need to run an MCP server locally. This is basically a
| proxy between an LLM and the proposed protocol.
|
| It's a nice proof of concept.
|
| And makes sense that the goal would be for LLM clients to adopt
| and support the standard natively. Then the proxy won't be
| necessary.
| koolala wrote:
| ChatGPT couldn't websocket connect to it directly in your
| browser?
| sublinear wrote:
| From https://modelcontextprotocol.io/docs/getting-started/intro
|
| > What can MCP enable? 1) Agents can access your Google Calendar
| and Notion, acting as a more personalized AI assistant. 2) Claude
| Code can generate an entire web app using a Figma design. 3)
| Enterprise chatbots can connect to multiple databases across an
| organization, empowering users to analyze data using chat. 4) AI
| models can create 3D designs on Blender and print them out using
| a 3D printer.
|
| Sure 1 and 3 make sense if they mean "summarize" and not
| "analyze", 4 maybe, but 2... Oh I don't know where to begin other
| than to say that even really smart humans have a very hard time
| dealing with that task based on a figma doc. Wouldn't it make
| more sense to generate the figma doc if they're already that
| awful to begin with?
| miguelspizza wrote:
| Hey glad to see Jason getting the credit he deserves! He was
| grokking WebMCP long before anyone else.
|
| If anyone wants to test out WebMCP, you can go to:
| https://webmcp.sh/ (this is a WebMCP server)
|
| With the MCP-B chrome extension (this is a WebMCP client):
| https://chromewebstore.google.com/detail/mcp-b-extension/dao...
|
| and use it to call WebMCP tools
| dogma1138 wrote:
| Sitemap.xml is back.
| pjmlp wrote:
| Never went away, we always have to deliver it somehow on our
| projects.
| mrasong wrote:
| Tried Claude's Chrome control feature before, it totally blew my
| mind.
|
| Didn't expect WebMCP to let you build custom MCP clients too.
| That's actually way more flexible!
| hhthrowaway1230 wrote:
| Good work! Needed this today
| saberience wrote:
| This seems like a security nightmare, a way to inject insecure
| content onto everyone's PC which can then automate actions
| executed with full user/admin privileges?
|
| Why on Earth would I want this?
| jasonjmcghee wrote:
| Entirely executing within the browser sandbox is one way to
| mitigate this. And that is the current scope of the official
| w3c proposal for this.
|
| https://github.com/webmachinelearning/webmcp
|
| I attempted to acknowledge the security implications and am not
| trying to push this as a product/service - this was just a
| proposal.
|
| Despite it being a proposal, I added token based authentication
| to mitigate potential abuse by forcing users to intentionally
| authenticate with a website before it can be used.
| brazukadev wrote:
| Why would I use that in place of asking the model to use an API?
| /s
| ngc6677 wrote:
| Also an other take here https://gitlab.com/sctlib/bmcp
| socketcluster wrote:
| My understanding is that this is essentially a hub for MCP
| servers?
|
| The current and potential benefits are:
|
| - Consistent authentication mechanism for all tools.
|
| - Ease of tool registration/deregistration.
|
| - Tool discovery.
|
| Main drawbacks are:
|
| - Trusting WebMCP npm package to run on your users' computers.
|
| - Trusting WebMCP with access to your site or platform's
| functionality. Kind of like OAuth?
|
| Does this sound right? Any other pros and cons versus integrating
| MCP server directly with specific LLMs?
| jasonjmcghee wrote:
| That's not how I'd describe it- it's not meant to centralize
| servers, it's the idea: maybe you don't need to build and
| distribute a separate downloadable thing for users to interact
| with your service/product/whatever via agent, and instead they
| continue to use your website via an appropriate interface for
| agents.
|
| (Here's another comment with an explanation
| https://news.ycombinator.com/item?id=45623782)
|
| The npm package is only there as the browser doesn't natively
| support the behavior (yet). Similarly MCP clients don't have
| built in support. So it's a bridge/proxy to demonstrate what
| could be done.
| socketcluster wrote:
| I think the centralization aspect sounds potentially very
| useful so I didn't mean it like a deal-breaker. I've been
| thinking it's a matter of time before someone figures out a
| good way to centralize MCP tools. This kind of thing could be
| huge; like the Google of MCP tools.
___________________________________________________________________
(page generated 2025-10-18 23:02 UTC)