Twenty-two lines added, two removed, one file. That's the entire diff for a pull request titled "Add consequentialHint to ToolAnnotations," merged into the WebMCP repository on September 3, 2026, commit 7b3f50f31848b529e69bedbbdf8da0edccba055f, authored by François Beaufort.
Last week we asked whether A2UI's rendering problem would be resolved by standards converging. This week the plumbing firmed up somewhere else — consequentialHint now exists as a named spec primitive in WebMCP — and the interesting question moved off convergence entirely. It is now: who decides what counts as consequential.
The trail starts on May 8, 2026, when Johann Hofmann opened issue #176, "Hint for reversible or consequential actions," out of a conversation with another contributor. The problem statement offers one clean example. Drafting an email isn't a read-only action, but it isn't something that commonly needs confirming either. Sending the email isn't reversible and should definitely be confirmed. readOnlyHint, already in the spec, was too blunt a map for that territory.
Drafting an email isn't a read-only action, but it isn't something that commonly needs confirming either.
Hofmann noted that MCP's destructiveHint "comes close but doesn't feel like it quite hits the same note — there are many possible non-deletion actions that users may take that warrant user confirmation." Victor Huang argued that destructive reads as delete or remove, and that consequential "can have clarity issues, and might end up being used too loosely." Mozilla's bvandersloot suggested the real gap might be Creates, which