Agentic AI Foundation coordination

[anssiko]

When we revise the charter the next time, look into formalizing our coordination with https://aaif.io/ for WebMCP and others with an interop opportunity.

Cc @dontcallmedom

[vthub]

Excellent! I see a few potential paths: we could either aim for a w3c liaison relationship to align charters, or start with a lighter joint task force focused specifically on browser-side technical compatibility (which can be either strict spec parity or a looser co-evolution model where we ensure compatibility of specs). I will work to get AAIF looped in to explore and start that conversation.

[anssiko]

Context: We just discussed this topic with @vthub (from Block, AAIF co-founder) and he volunteered to work with us on this coordination.

[vthub]

Started a thread in MCP discord to discuss potential working model between MCP and WebMCP.

My preliminary thoughts on how we can work with MCP / AAIF group, keeping it lightweight, flexible, yet ensuring checks and balances are in place as two specs grow and develop:

• Scope. On WebMCP, we explicitly enumerate what MCP spec sections apply to us -- not all MCP concepts are relevant in the browser context.
• Intent. We set and document an intent within Web ML working group to keep WebMCP aligned with MCP specification as its development progresses. By default, any proposal for MCP spec change should end up on the agenda for WebML working group meeting.
• Engagement. As needed, we use existing MCP and WebMCP communication channels to provide feedback / input on sections of interest: github, discord, mcp working group meetings.

Would love any feedback.

[anssiko]

Discussed this with @rigow who volunteered to review the proposal.

[rigow]

If we want to have a relation with AAIF, we need to agree with them on participation. If we just want to monitor and include things in our own work, we can do so, as MCP is developed under an MIT license, meaning we will need to give some attribution, but that's it.
So IMHO the aims above can be done without any formal relation to AAIF. If you have further aims beyond technical alignment, I would defer to @dontcallmedom for further strategic insight.

[anssiko]

Thanks @rigow for your review. I'll bring this topic for discussion in the group.

[domfarolino]

During the WebML call yesterday I asked what exactly the goal of our coordination was, and specifically if we were trying to institute a new formal bureaucracy or approval mechanism/etc. It sounds like no, which is nice! But I did have one thought/recommendation, drawn from experience in the WHATWG. All WHATWG spec PRs must be accompanied by implementation bugs, for each implementer of the spec. Most of the time this is just 3 browsers, but for a subset of APIs implemented by non-browsers, we file bugs against Node and Deno, since they're part of the same community. See https://github.com/whatwg/html/blob/main/.github/pull_request_template.md:

• Deno (only for timers, structured clone, base64 utils, channel messaging, module resolution, web workers, and web storage): …
• Node.js (only for timers, structured clone, base64 utils, channel messaging, and module resolution): …

The MCP spec will evolve in lots of ways that are not relevant to WebMCP, but a subset of choices made will be relevant to WebMCP. One example is webmachinelearning/webmcp#145 (comment), where we note that WebMCP is kind of like an outdated version of actual MCP without the title attribute in tool declarations.

Perhaps when MCP makes changes to things like this, e.g., tool declarations, schema support, or other things, they could have a process of filing an issue against WebMCP just so we're aware about it and can discuss whether we need to react. Thoughts? /cc @vthub