perf(query): semantic search creates new Memgraph connection per call instead of reusing

[vitali87]

Problem

In codebase_rag/tools/semantic_search.py, the semantic_code_search() function creates a new MemgraphIngestor context manager on every call rather than reusing the existing ingestor connection that the orchestrator already has. This adds unnecessary connection overhead per semantic search invocation.

Additionally, the code calls the private method ingestor._execute_query() directly instead of using the public fetch_all() API, which bypasses the public interface contract.

Expected Behavior

1. Reuse the existing Memgraph connection/ingestor instance instead of creating a new one per search call
2. Use the public fetch_all() method instead of the private _execute_query()

Affected Files

• codebase_rag/tools/semantic_search.py (lines 39-46)

[ChetanyaRathi]

Hi @vitali87, I'd like to take this. I've injected the orchestrator's existing MemgraphIngestor into semantic_code_search and get_function_source_code (same pattern was in both) via the create_*_tool factories and the call sites in main.py and mcp/tools.py, and switched the private _execute_query() calls to the public fetch_all(). Existing tests migrated + a new test asserting the injected connection is reused. PR incoming.
Thank you!