`search_metadata` and `semantic_search` MCP tools do not respect domain-based RBAC boundaries

[eeshan-deshpande]

Affected module

Backend

Describe the bug

Both MCP tools can return entities outside a caller's permitted domains, for two different reasons:

1. semantic_search never applies RBAC/domain filtering at all. It doesn't pass SubjectContext into the vector search path (OpenSearchVectorService / VectorSearchQueryBuilder), so results aren't scoped to the caller's role/policy regardless of settings.

2. search_metadata does pass SubjectContext, but domain filtering is silently skipped whenever the calling identity is a bot. SearchUtils.shouldApplyRbacConditions() requires !subjectContext.isBot() before applying RBAC conditions — so bot-authenticated tokens (the standard way to connect MCP) bypass domain scoping even with Search RBAC enabled. Confirmed with a real user PAT it can still leak, suggesting an unconditioned ViewAll (e.g. the default Organization Policy) is overriding the domain-scoped rule.

To Reproduce

1. Create two Domains (e.g. Finance, Marketing) with assets in each.
2. Create a role scoped to one domain only (hasDomain()), assign it to a test user/bot.
3. Query search_metadata and semantic_search via MCP with that identity's token.
4. Observe: assets from the other domain are still returned.

Expected behavior

Both tools should scope results to only the domains the caller is permitted to access, matching the behavior of the regular UI/API search.

OS

No response

Python version

No response

OpenMetadata version

1.12.5

OpenMetadata Ingestion package version

No response

Additional context

No response

Pre-submission checklist

• I searched for duplicate issues.
• I removed credentials, hostnames, emails, and other sensitive data from logs and config.