mcp-server-qdrant

What it does

Mcp-server-qdrant is the official Model Context Protocol server for Qdrant, a vector search engine. It exposes two tools: qdrant-store persists information (with optional metadata) to a Qdrant collection, while qdrant-find retrieves semantically relevant results via vector search. Unlike keyword-based retrieval, the server embeds text into high-dimensional vectors, enabling meaning-aware lookups. It supports both local file-based Qdrant databases and remote Qdrant servers, functioning as a semantic memory layer for LLM applications.

Who it's for

Developers building LLM applications that require persistent, searchable semantic memory. This includes AI agents tracking conversation history, question-answering systems grounded in document collections, and multi-turn workflows where semantic relevance outweighs exact text matching.

Common use cases

• Store and recall user context across sessions, allowing an LLM agent to retrieve prior conversations by semantic similarity
• Build a knowledge base that Claude queries during reasoning, filtering results by semantic relevance rather than keyword overlap
• Index conversation or audit logs semantically, enabling retrieval by conversational intent rather than exact phrasing

Setup pitfalls

• Cannot specify both QDRANT_URL and QDRANT_LOCAL_PATH; configure one or the other
• Embedding models download automatically on first startup; ensure network access and disk space for the default sentence-transformers/all-MiniLM-L6-v2 model
• COLLECTION_NAME environment variable is required unless overridden per request; misconfiguration fails silently
• Remote Qdrant servers with authentication enabled require QDRANT_API_KEY