fix: derive qdrant model params without private api (#60)

* fix: derive qdrant model params without private api

* Update mcp_plex/loader.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: Teagan42

AGENTS.md

@@ -27,6 +27,9 @@
   split into batches of five.
 - Qdrant upserts are batched and network errors are logged so large loads can
   proceed even when individual batches fail.
+- Qdrant model metadata is tracked locally in a supported mapping so the loader
+  no longer depends on the private `_get_model_params` helper removed in newer
+  clients.
 
 ## User Queries
 The project should handle natural-language searches and recommendations such as:

mcp_plex/loader.py

@@ -47,6 +47,17 @@
 _imdb_batch_limit: int = 5
 _qdrant_batch_size: int = 1000
 
+# Known Qdrant-managed dense embedding models with their dimensionality and
+# similarity metric. To support a new server-side embedding model, add an entry
+# here with the appropriate vector size and `models.Distance` value.
+_DENSE_MODEL_PARAMS: dict[str, tuple[int, models.Distance]] = {
+    "BAAI/bge-small-en-v1.5": (384, models.Distance.COSINE),
+    "BAAI/bge-base-en-v1.5": (768, models.Distance.COSINE),
+    "BAAI/bge-large-en-v1.5": (1024, models.Distance.COSINE),
+    "text-embedding-3-small": (1536, models.Distance.COSINE),
+    "text-embedding-3-large": (3072, models.Distance.COSINE),
+}
+
 
 async def _gather_in_batches(
     tasks: Sequence[Awaitable[T]], batch_size: int
@@ -62,6 +73,17 @@ async def _gather_in_batches(
     return results
 
 
+def _resolve_dense_model_params(model_name: str) -> tuple[int, models.Distance]:
+    """Look up Qdrant vector parameters for a known dense embedding model."""
+
+    try:
+        return _DENSE_MODEL_PARAMS[model_name]
+    except KeyError as exc:
+        raise ValueError(
+            f"Unknown dense embedding model '{model_name}'. Update _DENSE_MODEL_PARAMS with the model's size and distance."
+        ) from exc
+
+
 async def _fetch_imdb(client: httpx.AsyncClient, imdb_id: str) -> Optional[IMDbTitle]:
     """Fetch metadata for an IMDb ID with caching and retry logic."""
 
@@ -636,6 +658,7 @@ async def run(
             )
         )
 
+    dense_size, dense_distance = _resolve_dense_model_params(dense_model_name)
     if qdrant_url is None and qdrant_host is None:
         qdrant_url = ":memory:"
     client = AsyncQdrantClient(
@@ -647,7 +670,6 @@ async def run(
         https=qdrant_https,
         prefer_grpc=qdrant_prefer_grpc,
     )
-    dense_size, dense_distance = client._get_model_params(dense_model_name)
     collection_name = "media-items"
     created_collection = False
     if not await client.collection_exists(collection_name):

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mcp-plex"
-version = "0.26.27"
+version = "0.26.28"
 
 description = "Plex-Oriented Model Context Protocol Server"
 requires-python = ">=3.11,<3.13"

tests/test_loader_integration.py

@@ -6,6 +6,7 @@
 
 from qdrant_client.async_qdrant_client import AsyncQdrantClient
 from qdrant_client import models
+import pytest
 
 from mcp_plex import loader
 
@@ -97,3 +98,20 @@ def test_run_upserts_in_batches(monkeypatch):
     asyncio.run(_run_loader(sample_dir))
     assert CaptureClient.upsert_calls == 2
     assert len(CaptureClient.captured_points) == 2
+
+
+def test_run_raises_for_unknown_dense_model():
+    sample_dir = Path(__file__).resolve().parents[1] / "sample-data"
+
+    with pytest.raises(ValueError, match="Unknown dense embedding model"):
+        asyncio.run(
+            loader.run(
+                None,
+                None,
+                None,
+                sample_dir,
+                None,
+                None,
+                dense_model_name="not-a-real/model",
+            )
+        )

tests/test_loader_unit.py

@@ -7,6 +7,7 @@
 
 import httpx
 from qdrant_client import models
+import pytest
 
 from mcp_plex import loader
 from mcp_plex.imdb_cache import IMDbCache
@@ -22,6 +23,7 @@
     _load_imdb_retry_queue,
     _persist_imdb_retry_queue,
     _process_imdb_retry_queue,
+    _resolve_dense_model_params,
     resolve_tmdb_season_number,
 )
 from mcp_plex.types import TMDBSeason, TMDBShow
@@ -482,3 +484,14 @@ async def upsert(self, collection_name: str, points, **kwargs):
     monkeypatch.setattr(loader, "_qdrant_batch_size", 1)
     asyncio.run(loader._upsert_in_batches(client, "c", points))
     assert client.calls == 3
+
+
+def test_resolve_dense_model_params_known_model():
+    size, distance = _resolve_dense_model_params("BAAI/bge-small-en-v1.5")
+    assert size == 384
+    assert distance is models.Distance.COSINE
+
+
+def test_resolve_dense_model_params_unknown_model():
+    with pytest.raises(ValueError, match="Unknown dense embedding model"):
+        _resolve_dense_model_params("not-a-real/model")