OpenInquiry
OpenInquiry now demonstrates the core idea. Next, it needs testing on independent publisher sites, a simple implementation path, and an open process shaped by publishers, readers, and agent developers.
01
Start with an installable package that connects OpenInquiry to the search, sign-in, access, and content systems already on a publisher’s site. Adding WebMCP Site Tools should not require rebuilding those systems or creating a separate AI product.
02
Let each publisher choose what to return: full text, a complete section, an excerpt, a summary, basic details, a link, or no additional content. OpenInquiry should not impose one character limit. Publishers should set limits by resource and permitted use.
03
WebMCP gives publishers a controlled tool path, but an agent-capable browser may still read the visible page directly. Test a browser-enforced mode in which protected text reaches the agent only through declared Site Tools while the reader keeps normal access. Any credential used for that decision must be signed, scoped, and revocable, with explicit claims about retention, training, onward sharing, and deletion. WebMCP does not provide this guarantee today, and self-reporting is not enough.
04
Publish proposed changes and the reasoning behind them. Governance should include publishers, readers, agent developers, and independent standards participants.