Replies: 2 comments
|
Thanks for raising this. Digging into it, we found the discoverability issue was actually at the schema level. The listing mode (empty Q1 (relax Q2 (separate Routing validated against LM Studio + Qwen3.5 9B across 5 scenarios. With the new signature the model adopts the cleaner form, dropping PR: #1004 |
Uh oh!
There was an error while loading. Please reload this page.
ha_search_entities currently requires at least one search parameter (query, domain, device_class, or area_id). This is likely intentional — returning all entities unconstrained on large instances is a real context-size risk.
However, there are use cases where an agent needs a flat, filterable list of all entities — e.g. to audit a domain or build a picker — without knowing a search term upfront. Other MCP servers in the ecosystem address this with a dedicated
list_entities(domain?, detailed?, fields?)tool.Two questions for the maintainers:
ha_get_overview(detail_level="full", domains="x")the intended path for this use case, or would relaxingha_search_entitiesto allow domain-only filtering without a query make sense?ha_list_entitiestool, or does that conflict with the tool-count philosophy?Happy to implement whichever direction is preferred, or close this if ha_get_overview is considered sufficient.
All reactions