What it does

The mcp-server-qdrant provides a semantic memory interface to Qdrant, a vector database designed for high-performance similarity search. It exposes two core tools — qdrant-store persists information with optional metadata into the vector database, and qdrant-find retrieves semantically relevant results through vector similarity search. The server automatically handles embedding generation via configurable embedding models and manages all communication with Qdrant instances, whether remote or local.

Who it's for

Developers building LLM applications that need persistent semantic memory or vector retrieval capabilities. Primary use cases span AI assistants requiring long-term context retention, retrieval-augmented generation (RAG) systems, and knowledge base applications.

Common use cases

Store and retrieve conversation histories via semantic search
Build RAG pipelines that fetch relevant documents for LLM context
Maintain multi-turn agent memory across sessions
Index and search application-specific knowledge bases

Setup pitfalls

Cannot use both QDRANT_URL (for a remote server) and QDRANT_LOCAL_PATH (for a local embedded database) at the same time — choose one approach
Remote Qdrant instances require QDRANT_API_KEY authentication; ensure the key is valid before deploying
Embedding model defaults to sentence-transformers/all-MiniLM-L6-v2 but may need customization for domain-specific or specialized vector representations
Read-only mode (QDRANT_READ_ONLY=true) disables the store tool entirely, limiting the server to retrieval only