Decision needed
Should the docs.plus connector ship an interactive UI widget, or stay text-only?
MCP Apps let a tool render an interactive view inside the conversation, instead of returning text. The same widget renders in Claude and, per the SDK's own documentation, in ChatGPT.
What we know
Research on 2026-09-04 covered this and did not task it. Text tools ship first, and a widget only earns its place once a text tool proves too thin.
Two places where a widget might beat text:
- A document picker.
find_documents returns a list. A searchable picker beats a text list once a person has many documents.
- A diff view. Once the write tools land, showing a proposed change visually beats describing it.
What to weigh
A widget is a second front end: HTML served as an MCP resource, in a sandboxed iframe, with its own content security policy where every origin must be declared. That is real surface area for a small team.
Answer this after #226 and #228 have served real users. The evidence should be a text tool that people find awkward, not a guess.
Acceptance
Decision needed
Should the docs.plus connector ship an interactive UI widget, or stay text-only?
MCP Apps let a tool render an interactive view inside the conversation, instead of returning text. The same widget renders in Claude and, per the SDK's own documentation, in ChatGPT.
What we know
Research on 2026-09-04 covered this and did not task it. Text tools ship first, and a widget only earns its place once a text tool proves too thin.
Two places where a widget might beat text:
find_documentsreturns a list. A searchable picker beats a text list once a person has many documents.What to weigh
A widget is a second front end: HTML served as an MCP resource, in a sandboxed iframe, with its own content security policy where every origin must be declared. That is real surface area for a small team.
Answer this after #226 and #228 have served real users. The evidence should be a text tool that people find awkward, not a guess.
Acceptance