Improve search relevance with scoring and word-boundary matching #2
Description
Summary
Search returned too many irrelevant results due to substring matching and missing field coverage.
Changes
- Relevance scoring: namespace (20), toolName (15), tags (12), schemaName (8), description (5)
- Multi-word query support with AND logic
- Word-boundary matching instead of substring (eliminates false positives like "tokens" matching "ens")
- Tags and schemaName propagated from schemas into tool objects
- Output extended with
namespace,tags,scorefields hintfield for >10 matches and zero results