Hosting official polyfill for WebMCP

[beaufortfrancois]

Varying native integration timelines across Safari, Firefox, Chromium-based browsers, and others create a distinct support gap. To address this platform deficiency, early adopters are currently independently creating custom polyfills for WebMCP:

• https://github.com/jeanbaptiste/webmcp-auto-ui/blob/main/packages/core/src/polyfill.ts
• https://github.com/WebMCP-org/npm-packages/tree/main/packages/webmcp-polyfill
• https://github.com/GoogleChromeLabs/webmcp-tools/blob/main/demos/shared/webmcp-polyfill.js

Consolidating these under the official webmachinelearning org would provide a clear, authoritative source of truth for the community.

Would you be open to hosting an official polyfill in this org like we did for TypeScript types under a new dedicated repo named https://github.com/webmachinelearning/webmcp-polyfill for instance?

[domfarolino]

Totally agreed that we should do this! @anssiko, how does this sound to you? I think it'd be a great way to onboard developers onto the project in an official capacity.

[anssiko]

I have this slated for discussion today.

[anssiko]

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.

[igrigorik]

This is great, excited to see interest in maintaining an official polyfill!

I'm a fan of @mcp-b/webmcp-polyfill. It's MIT, already on the current document.modelContext shape, and the WebIDL work is careful in ways that are easy to get wrong.

As an FYI, I also rolled a separate implementation in AgentBoard. It's written for a browser extension rather than for site authors — injected into pages so the extension can consume tools from any site — so it's deliberately a subset: same-document imperative core, no declarative forms or cross-origin. It's conformance-tested against pinned upstream WPT via this harness.