As a co-maintainer of XMLUI I’m acutely delicate to brokers fumbling as they construct XMLUI apps. When one thing that needs to be a no brainer isn’t, as a result of the MCP search didn’t discover the reply it ought to have, I file a problem after which direct an agent to put in writing the lacking doc.
That new doc is a testable repair. For instance, I requested Claude to seek out its failed searches for the select-to-radio conversion within the MCP log. The response:
Two how-to searches, each at 21:12, each misses:
“radio group mutually exclusive choice change value form input” — high hits: use-a-custom-formitem, prefill-a-form-from-an-api-response, use-regex-validation-in-formitem. Form-adjacent noise, zero radio content material.
“choose one of two options horizontal selector” — the rephrase moments later, attacking from the UX facet as an alternative of the part facet. Also nothing.
The failing searches had returned solely weak matches with a high relevance rating round 0.7. After including the brand new doc I requested it to repeat the search and examine. The new doc surfaced at #1 with a 4.2 rating, six instances larger than the rest. (Because the MCP server can pin the model of the docs it consults, direct A/B comparability is feasible.) I additionally tried a handful of artificial queries, like “radio buttons for a small set of options”. These confirmed the end result.
