[ojaswa072]

Problem

MCPToolProvider._log_activity performed a synchronous SQLAlchemy db_session() write directly on the async event loop. This is a hot path, it fires on every MCP tool call (pre-call, success, and failure paths) plus once per server during connect(). Under concurrent sessions this serializes all MCP tool executions through the event loop, adding DB round-trip latency to every agent action and contributing to connection pool pressure.

Fix

• Extract the blocking SQLAlchemy body into _log_activity_sync (private sync helper).
• Make _log_activity an async def that dispatches to asyncio.to_thread(_log_activity_sync, ...), keeping the event loop unblocked.
• await all five call sites: one in connect(), three in call_mcp_tool().
• Add import asyncio.

Behaviour

No change to the audit log, writes still happen, just on a thread-pool worker. The fix is non-breaking; _log_activity_sync is a private method and no external callers exist.