mcp_/manager: start_server_sync's no-loop fallback reports RUNNING without starting anything; hasattr-lazy task registries; duplicated background-task plumbing

File: code_puppy/mcp_/manager.py - start_server_sync / stop_server_sync (~lines 576-700)

Severity: Low-Medium (correctness / Zen)

1. No-event-loop fallback lies about state. In the except RuntimeError: branch of start_server_sync, the server is merely enable()d but the status tracker is set to ServerState.RUNNING and a start time is recorded - no process was started and no health check ran:

except RuntimeError:
    # No async loop, just enable the server
    managed_server.enable()
    self.status_tracker.set_status(server_id, ServerState.RUNNING)  # <-- nothing is running

The async path correctly distinguishes STARTING -> RUNNING. The dashboard / /mcp listing will show RUNNING for a server whose subprocess never launched. Suggested: use a distinct state (ENABLED/STOPPED with enabled=True, or STARTING) and let the first actual use transition it.

2. Lazy attribute creation via hasattr - _pending_start_tasks / _pending_stop_tasks are created on first use:

if not hasattr(self, "_pending_start_tasks"):
    self._pending_start_tasks = {}

Initialize both in __init__ (explicit is better than implicit); the cleanup_task callbacks also re-check hasattr for the same reason. This pattern appears 4 times in the file.

3. The start_server_background / stop_server_background closures + task bookkeeping are copy-pasted between the two sync wrappers - extract a _schedule_background(coro_fn, registry_dict, name) helper.

Filed by Zen Reviewer B (code-puppy-60635a)

Start in code_puppy/mcp_/manager.py around start_server_sync and stop_server_sync (~lines 576-700), then inspect init and the four hasattr checks for pending task registries. Trace the no-event-loop state update and the duplicated background-task closures. Done means state reflects whether a process actually started, registries are initialized explicitly, and shared scheduling behavior remains consistent.