# [Bug] MCP gateway: every tool call fires a false-positive `llm_requests_hanging` alert — synthetic MCP data has no `litellm_call_id` and MCP completion never writes `request_status`

### What happened

With alerting enabled (`alert_types` including `llm_requests_hanging`, `alerting_threshold: 300`), **every MCP gateway tool call produces a "Requests are hanging - 300s+ request time" alert ~300s later**, with `Request Model: mcp-tool-call`, `API Base: None`, and empty key/team alias — even though the tool call completed successfully within seconds.

In our deployment a scheduled job makes a handful of MCP tool calls every 6 hours; the job finishes in <2 minutes, yet we receive a hanging alert exactly 300s (+ checker tick) after its last MCP call, every single run. The alert timestamps are metronomic: `HH:06:46` for a job whose last MCP call lands at `~HH:01:46`.

### Root cause (traced in v1.91.0; still present on `main`)

Three code facts combine into a false positive **by construction**:

1. **Registration**: `MCPServerManager` runs the standard proxy pre-call hook for every tool call (`proxy/_experimental/mcp_server/mcp_server_manager.py`, `pre_call_hook(..., call_type=CallType…