Hosting official TypeScript types for WebMCP

[beaufortfrancois]

As the WebMCP ecosystem grows, having standardized, easily accessible TypeScript types will be highly beneficial for developer adoption. I'm opening this issue to ask whether the webmachinelearning organization would be the right place to host the official TypeScript type definitions for WebMCP.

Currently, types are a bit fragmented in the wild:

• webmcp-tools/tree/main/demos/shared/types
• WebMCP-org/npm-packages

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

Questions for the maintainers:

1. Would you be open to hosting the official TypeScript types in the https://github.com/webmachinelearning/ org?
2. If yes, should they live in this repo or a new dedicated one?
3. How can we best help consolidate the existing types and move them here?

[tomayac]

FWIW, built-in AI types are on DefinitelyTyped.

[OrKoN]

DefinitelyTyped is generally a better place for types from the client perspective.

[anssiko]

This community hosts https://github.com/webmachinelearning/webnn-types so as an organization we're set up for more types. To make contributing easier, a separate repo is preferred, but also a directory in this repo would work. If a new repo sounds interesting, I'd create it for you and expect to see interested maintainer(s) willing to take ownership of it.

[beaufortfrancois]

Since this API is still in incubation, I'd prefer using a repository like https://github.com/webmachinelearning/webmcp-types for now. Once the API stabilizes across browsers, the next step would be moving those to DefinitelyTyped as a @types/webmcp package. I'm happy to discuss this further.

[MiguelsPizza]

+1. I maintain the types here: WebMCP-org/npm-packages. My version of the lib goes quite a bit into the type inference side of things (infers the type of execute args based on input schema).

This is nice for DX but does add some functionality & assumptions into the api that are not actually enforced yet (that the input args will adhere to the input schema for example).

@beaufortfrancois WDYT? Is type inference something the official type package should concern itself with (happy to help maintain the new package if so)? Or should I continue to maintain this separately for those who want type inference?