Fix lifespan context handling for SSE connections

mcp_plex/server.py

@@ -6,7 +6,6 @@
 import inspect
 import json
 import os
-from contextlib import asynccontextmanager
 from typing import Annotated, Any, Callable
 
 from fastapi import FastAPI
@@ -57,12 +56,18 @@ def __init__(
             https=self.settings.qdrant_https,
         )
 
-        @asynccontextmanager
-        async def _lifespan(app: FastMCP):  # noqa: ARG001
-            try:
-                yield
-            finally:
-                await self.close()
+        class _ServerLifespan:
+            def __init__(self, plex_server: "PlexServer") -> None:
+                self._plex_server = plex_server
+
+            async def __aenter__(self) -> None:  # noqa: D401 - matching protocol
+                return None
+
+            async def __aexit__(self, exc_type, exc, tb) -> None:  # noqa: ANN001
+                await self._plex_server.close()
+
+        def _lifespan(app: FastMCP) -> _ServerLifespan:  # noqa: ARG001
+            return _ServerLifespan(self)
 
         super().__init__(lifespan=_lifespan)
         self._reranker: CrossEncoder | None = None

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mcp-plex"
-version = "0.26.29"
+version = "0.26.30"
 
 description = "Plex-Oriented Model Context Protocol Server"
 requires-python = ">=3.11,<3.13"

tests/test_server.py

@@ -208,8 +208,17 @@ def test_rest_endpoints(monkeypatch):
 
 def test_server_lifespan_context(monkeypatch):
     with _load_server(monkeypatch) as module:
+        closed = False
+
+        async def fake_close() -> None:
+            nonlocal closed
+            closed = True
+
+        monkeypatch.setattr(module.server, "close", fake_close)
+
         async def _lifespan() -> None:
             async with module.server._mcp_server.lifespan(module.server):
                 pass
 
         asyncio.run(_lifespan())
+        assert closed is True