Add OpenAI-compatible embedding provider support

pyproject.toml

@@ -10,6 +10,7 @@ dependencies = [
     "qdrant-client>=1.12.0",
     "pydantic>=2.10.6",
     "fastmcp>=2.7.0",
+    "openai>=1.109.1",
 ]
 
 [build-system]

src/mcp_server_qdrant/embeddings/factory.py

@@ -13,5 +13,11 @@ def create_embedding_provider(settings: EmbeddingProviderSettings) -> EmbeddingP
         from mcp_server_qdrant.embeddings.fastembed import FastEmbedProvider
 
         return FastEmbedProvider(settings.model_name)
+    elif settings.provider_type == EmbeddingProviderType.OAI_COMPAT:
+        from mcp_server_qdrant.embeddings.oai_compat import OAICompatProvider
+
+        return OAICompatProvider(settings.model_name,
+                                settings.oai_compat_endpoint,
+                                settings.oai_compat_api_key)
     else:
         raise ValueError(f"Unsupported embedding provider: {settings.provider_type}")

src/mcp_server_qdrant/embeddings/oai_compat.py

@@ -0,0 +1,48 @@
+from mcp_server_qdrant.embeddings.base import EmbeddingProvider
+from openai import AsyncOpenAI
+import asyncio
+
+
+class OAICompatProvider(EmbeddingProvider):
+    def __init__(self, model_name: str, endpoint_url: str, api_key: str, vec_size: int | None = None) -> None:
+        self.model_name = model_name
+        self.base_url = endpoint_url
+        self.key = api_key
+        self.oai_client = AsyncOpenAI(base_url=self.base_url, api_key=self.key)
+        self._vec_size = self.fetch_vec_size() if vec_size is None else vec_size
+
+    async def embed_documents(self, documents: list[str]) -> list[list[float]]:
+        response = await self.oai_client.embeddings.create(
+            input=documents,
+            model=self.model_name
+        )
+        results = [embed.embedding for embed in response.data]
+        return results
+
+    async def embed_query(self, query: str) -> list[float]:
+        response = await self.oai_client.embeddings.create(
+            input=query,
+            model=self.model_name
+        )
+        results = response.data[0].embedding
+        return results
+
+    def get_vector_name(self) -> str:
+        return f"oai_compat-{self.model_name}"
+
+    def get_vector_size(self) -> int:
+        return self._vec_size
+
+    def fetch_vec_size(self):
+        """Retrieves the embedding vector dimensionality by sending a minimal inference request to the endpoint.
+        Since the OpenAI-compatible API does not expose model metadata detailing vector dimensions, this method
+        uses a dummy input to obtain a sample embedding and derives the dimension count from its structure.
+        To avoid unnecessary latency and cost during initialization, it is recommended to explicitly provide
+        the vector size via the `vec_size` parameter when the dimension is known."""
+        try:
+            loop = asyncio.get_running_loop()
+            fut = asyncio.run_coroutine_threadsafe(self.embed_query("dummy"), loop)
+            result = fut.result()
+        except RuntimeError:
+            result = asyncio.run(self.embed_query("dummy"))
+        return len(result)

src/mcp_server_qdrant/embeddings/types.py

@@ -3,3 +3,4 @@
 
 class EmbeddingProviderType(Enum):
     FASTEMBED = "fastembed"
+    OAI_COMPAT = "oai_compat"

src/mcp_server_qdrant/settings.py

@@ -47,6 +47,21 @@ class EmbeddingProviderSettings(BaseSettings):
         validation_alias="EMBEDDING_MODEL",
     )
 
+    oai_compat_endpoint: str = Field(
+        default="https://api.openai.com/v1",
+        validation_alias="OAI_COMPAT_ENDPOINT"
+    )
+
+    oai_compat_api_key: str = Field(
+        default="",
+        validation_alias="OAI_COMPAT_API_KEY"
+    )
+
+    oai_compat_vec_size: int | None = Field(
+        default=None,
+        validation_alias="OAI_COMPAT_VEC_SIZE"
+    )
+
 
 class FilterableField(BaseModel):
     name: str = Field(description="The name of the field payload field to filter on")