W3C

– DRAFT –
WebML CG Teleconference – 20 August 2026

20 August 2026

Attendees

Present
Alex_Nahas, Anssi_Kostiainen, Brandon_Walderman, Dominic_Farolino, Ehsan_Toreini, Guru_Singh, Guy_Bary, Iris_Johnson, Jason_McGhee, Johann_Hofmann, Juan_Fernandez, Julia_Pagnucco, Julien_Bataille, Kannadasan_Saminathan, Mark_Foltz, Reilly_Grant, Sarah_Drasner, Saron_Yitbarek, Sushanth_Rajasankar, Thomas_Steiner, Victor_Huang, Yoav_Weiss
Regrets
-
Chair
Anssi
Scribe
Anssi, anssik

Meeting minutes

Anssi: as a reminder, we'll use IRC-based queue management in this meeting:

https://www.w3.org/guide/meetings/zakim.html#speakerqueue

Anssi: to suggest agenda topics, use Agenda+ label, e.g.:

Agenda+

Anssi: please welcome our latest new participants
… Mikhail Klimenko from NVIDIA
… Julien Bataille from Rakuten Group
… Vladimir Edouard from University of North Carolina at Chapel Hill
… Yaoming Liu from Huawei
… Amit Varia from Google
… Michael Lorenz from Karlsruhe Institute of Technology
… Sulaiman Alshammari from Gratech
… Jay Mok from PayPal
… Nicolas Couture, Carson Jozefzoon, Shoury Sharma, Felix Lind, Kannadasan Saminathan, Junjie Qiu and Yifan Chen as individual contributors
… welcome all!

Announcements

TPAC 2026 F2F

Anssi: WebML CG F2F confirmed for Tuesday 27 Oct 2026, in Dublin Ireland
… second day of the TPAC 2026 week 26-30 Oct

TPAC 2026 site

Anssi: a few logistics things to handle, please register for TPAC 2026 and book your travel and accommodation as soon as you can
… register latest by 01 October 2026:

TPAC 2026 registration

Anssi: discount rate provided at the meeting hotel:

TPAC 2026 accommodation

Anssi: our F2F Agenda is a joint GH issue with the WebML WG to allow the community chime in more easily with suggestions
… we can continue to label Agenda+ interesting issues and consider these candidates for discussion at the F2F

F2F agenda

<gb> Issue 39 WebML WG/CG F2F Agenda - TPAC 2026 (Dublin, Ireland) (by anssiko)

Anssi: the Working Group meets on Monday, 26 October 2026, right before the Community Group meeting
… I encourage all participants to attend both days to get the full picture of the WebML community

Anssi: questions, comments?

E-commerce for Humans and AI Agents workshop

Anssi: this W3C workshop aims to share experience of creating content with AI Agents with ecommerce as a particular focus
… the event takes place Sep 8-9, 2026 in Zurich, Switzerland, Google hosts
… workshop agenda has been published:

https://www.w3.org/2026/ecommerce-agents/agenda.html

Anssi: the event is now fully booked, but if you are interested in attending, please reach out to me and I can help you get on the waitlist
… more information about the workshop is available on the W3C's site:

https://www.w3.org/2026/ecommerce-agents/

WebMCP polyfill

Repository: webmachinelearning/webmcp

Anssi: issue #252

<gb> Issue 252 Hosting official polyfill for WebMCP (by beaufortfrancois)

Anssi: a proposal to host official polyfill for WebMCP in the webmachinelearning GH org

Anssi: there's a history for hosting polyfills, this is a good fit
… prior we hosted webmachinelearning/webnn-polyfill that served as a useful library while native implementations were still in development, we deprecated it when the native implementations matured
… share your feedback via the issue, we will make a decision in our next meeting

RESOLUTION: The group plans to host an official polyfill for WebMCP in the webmachinelearning GitHub organization. Adoption earliest by our next meeting, feedback via issue #252.

Translator and Language Detector APIs

Repository: webmachinelearning/translation-api

Security considerations for HTML translation

Anssi: issue #78

<gb> Issue 78 Document security considerations for HTML translation (by reillyeon) [Agenda+]

Anssi: Mozilla reviewed the API and suggested the way how HTML fragments are translated needs careful consideration to avoid security issues
… per Reilly's comment translation models expect text, not HTML as input
… possible vector for an injection attack

Reilly: this issue is outcome of time spent trying to respond to issue #60 around whether we could support translating HTML, still unsure whether we can spec this as an API or leave this to developers

<gb> Issue 60 Input formats for translation and language detection (by tomayac)

Reilly: took a look at Firefox whole page translation implementation, found it complicated
… proposing we respond to Mozilla's concerns and document that developers may implement translation along the lines of code snippet provided in the issue

Ehsan: I responded to the issue and support Reilly's proposed approach
… issue #60 has low-hanging fruit for consensus

<gb> Issue 60 Input formats for translation and language detection (by tomayac)

Ehsan: I tried to categorize related issues and proposed possible next steps

Anssi: are you volunteering to update the security considerations?

Ehsan: I'm working on that already

RESOLUTION: The group resolves to document the potential injection attack vector in the security considerations section of the Translator and Language Detector APIs specification. (issue #78)

<gb> Issue 78 Document security considerations for HTML translation (by reillyeon) [Agenda+]

Prompt API

Repository: webmachinelearning/prompt-api

Tool calling, extended thinking, and multimodality

Anssi: issue #209

<gb> Issue 209 Extend the Message format to support tool calling, extended thinking, and multimodality (by nico-martin) [Agenda+]

Anssi: for this issue, I'd want us to understand what gaps there are in the current multimodal and tool call support
… the issue was opened by Nico from Hugging Face, who is implementing a similar API shape for Transformers.js

Reilly: we haven't looked at tool calling lately too much, my recollection is we had something reasonable in this open-loop proposal
… trying to understand the gap with Nico's API implemented in Transformers.js

Sarah: should we defer this for our next meeting?

Reilly: when implementers are ready to think about this we should revisit this

Sarah: we will come to this by the next meeting

<reillyg> explainers-by-googlers/vector-store-api

WebMCP

Repository: webmachinelearning/webmcp

Anssi: I made a few changes to today's WebMCP agenda
… removed issues #186 and #227 that are fixed and closed, added issue #227

<gb> Issue 227 Tool discovery should not be limited to a single traversable navigable (by domfarolino) [Agenda+]

<gb> Issue 186 WebMCP "real-user measurement" (by yoavweiss) [Agenda+]

Anssi: great work by getting these issues addressed ahead the meeting

Wide review

Anssi: in our last meeting we initiated wide review of WebMCP with the TAG, Privacy and Security groups
… first, let's do a debrief on TAG, Privacy and Security reviews to bring the group up to date
… TAG has not yet delivered its review response
… TAG had a discussion on 2026-08-05 with WebML CG participants who were available at that time

2026-08-05 TAG meeting

Anssi: Dominic shared helpful notes in the TAG review issue:

w3ctag/design-reviews#1238 (comment)

<gb> Issue 1238 Incubation: WebMCP (by marcoscaceres) [Progress: in progress] [Review type: CG early review] [Venue: WebML CG] [Missing: Multi-stakeholder support] [Topic: Machine Learning] [Focus: API design (pending)] [Focus: Web architecture

<gb> … (pending)] [Focus: Security (pending)] [Focus: Privacy (pending)] [Focus: Accessibility (pending)] [Focus: Internationalization (pending)]

Anssi: per the meeting discussion, the TAG is still building understanding of WebMCP and its surrounding ecosystem
… I want to thank those who participated for diligently answering the questions and explaining the problem space WebMCP is addressing
… the realization that "more HTML semantics" is a different, complementary, problem space, helps clarify the scope of WebMCP

Dominic: good summary, one of the biggest hang ups is around security, folks not familiar with WebMCP can confuse the proposal with some other features, same-origin policy, mixing data across tabs, we're working hard with Threat Modeling folks at W3C to clarify the threat models for WebMCP and MCP are different
… we're making progress in building shared understanding on these issues

Anssi: did TAG share any estimate when the response would be delivered?

<Ehsan> the discussions in TAG still continues, Marcos already mentioned that today in the issue thread

Dominic: no definite dates shared

Ehsan: TAG discussions are still ongoing, the July discussion was not official TAG position

Dominic: if the TAG wants to have more discussion, we're happy to meet them in future breakouts
… Marcos not being present in the July breakout meant it was of reduced utility

Ehsan: Marcos initiated that discussion but was not in that meeting himself unfortunately

Anssi: privacy review request has been assigned to privacy reviewers and the spec was introduced on the privacy group's July meeting, no official response yet

w3cping/privacy-request#213

<gb> Issue 213 WebMCP 2026-06-25 > 2026-08-01 (by domfarolino) [WD] [pending] [REVIEW REQUESTED]

https://github.com/w3c-cg/threat-modeling/blob/agentic-ai-web-browsers-rewrite/meetings/2026-07-14_minutes.md

Anssi: security groups have had two sessions in July to discuss WebMCP, with some of our group participants attending

w3c/security-request#135

<gb> Issue 135 WebMCP 2026-06-25 > 2026-08-01 (by domfarolino) [REVIEW REQUESTED] [WD]

2026-07-07 Security IG

2026-07-14 Threat Modeling CG

Anssi: Simone is working on the WebMCP threat modeling exercise

Johann: Simone recently shared a draft for threat modeling, there's progress on this

Anssi: in the second phase of WebMCP wide review, we reach out to the web internationalization (i18n) and web accessibility (a11y) experts
… I'd propose we initiate wide reviews with the i18n and a11y groups in the coming few weeks
… I will coordinate this with the chairs of those groups
… we've already engaged with these groups via our GH issues:

i18n-tracker

a11y-tracker

Anssi: we will refer to these issues in our review requests

Anssi: in addition, both the groups provide checklists to catch the most common i18n and a11y issues:

a11y checklist

i18n checklist

Anssi: we're ought to validate WebMCP against these checklists to ensure we've covered the most important issues in our review process
… in my experience, many questions are geared toward HTML, and may not apply to APIs such as WebMCP
… in these cases we simply provide a rationale for why they don't apply, if not straightforward
… questions, comments, concerns?

Sarah: I know a few people who could help with a11y review

Kannadasan: I've worked in i18n space and can help a bit

Anssi: I will put interested folks willing to contribute in contact with i18n and a11y groups

RESOLUTION: Initiate wide review of WebMCP with the i18n and a11y groups along with collateral materials.

Developer tooling

Anssi: issue #207

<gb> Issue 207 Support developer tooling (by jackfranklin) [Agenda+]

Anssi: I'd like to understand key use cases and requirements for developer tooling wrt WebMCP
… and whether those use cases warrant providing a mechanism to distinguish between developer tools and user-facing tools as proposed
… two use cases are explored in the issue are:
… debugging web app's flickering UI
… debugging web app's state

Anssi: are we converging on the idea of a "devtoolstooldiscovery" event to help developers discover tools that are available for debugging purposes?

Sarah: for developer tooling, we're finding for large-scale apps it is good to have some scoping for tools, we could defer the cascade for tools, need some scoping to plugin Puppeteer and Playwright-style observability tools

Dominic: I think we'd want to standardize developer member in annotations, to be able to distinguish developer tools from non-developer tools

<domfarolino> +1

RESOLUTION: Add a debugging member to Annotation dictionary. (issue #207)

Real-user measurement

Anssi: issue #186

<gb> Issue 186 WebMCP "real-user measurement" (by yoavweiss) [Agenda+]

Anssi: this issue is to discuss how RUM techniques apply to WebMCP
… Yoav wants to find a solution that answers the questions:
… "Is my WebMCP implementation working well at scale?"
… "Did the user achieve what they were trying to do?"

Anssi: are these the core questions?

Yoav: we broke these down to sub questions in a follow up discussion
… some things we need to do are instrumentable
… with toolactivated event
… some other things developers need, e.g. what agents are users using and with what agents we should test with?
… agents may feel confused, was that a good experience for the user regardless?
… I don't know if we can provide these signals to the developer in the granularity required
… e.g. some timestamps could be useful

Dominic: we talk about this internally, my hunch is whether the agent succeeds in representing its user's task, this is something sites want to know even outside WebMCP
… I imagine this taking place outside WebMCP, site could register reporting URL
… WebMCP would be represented in that
… would like to start with something broader and let WebMCP feed into that
… Yoav, how do you think about that?

Yoav: we'd perhaps need some UA String -style mechanism to understand the user's agent
… disconnected from session, developers would need to setup collection infrastructure
… we can have various agent interaction metrics and WebMCP support will improve them, tool calls that are instrumentable

Yoav: let's continue discussion in the issue, reporting API outside WebMCP is interesting, let's explore straw proposals in this space

Summary of resolutions

  1. The group plans to host an official polyfill for WebMCP in the webmachinelearning GitHub organization. Adoption earliest by our next meeting, feedback via issue #252.
  2. The group resolves to document the potential injection attack vector in the security considerations section of the Translator and Language Detector APIs specification. (issue #78)
  3. Initiate wide review of WebMCP with the i18n and a11y groups along with collateral materials.
  4. Add a debugging member to Annotation dictionary. (issue #207)
Minutes manually created (not a transcript), formatted by scribe.perl version 248 (Mon Oct 27 20:04:16 2025 UTC).

Diagnostics

Succeeded: s/propose/proposing

Succeeded: s/prefix/member

Maybe present: Anssi, Dominic, Ehsan, Johann, Kannadasan, Reilly, Sarah, Yoav

All speakers: Anssi, Dominic, Ehsan, Johann, Kannadasan, Reilly, Sarah, Yoav

Active on IRC: anssik, domfarolino, Ehsan, Mark_Foltz, reillyg