mcp-server-qdrant

What it does

An official MCP server for Qdrant, a vector search engine, that acts as a semantic memory layer for AI applications. It exposes two tools: qdrant-store to persist text with optional metadata into a vector collection, and qdrant-find to retrieve contextually similar information via semantic search. The server supports both local Qdrant instances (via QDRANT_LOCAL_PATH) and remote deployments (via QDRANT_URL and QDRANT_API_KEY). Text is embedded using FastEmbed and the default model sentence-transformers/all-MiniLM-L6-v2, configurable via environment variables.

Who it's for

Backend engineers and AI developers building Claude-powered applications that need long-term, semantically searchable memory. Typical users: teams building persistent AI agents, chat applications requiring context retrieval beyond a single conversation window, or knowledge systems where keyword search is insufficient.

Common use cases

• Store and retrieve conversation history by semantic relevance rather than recency
• Maintain a persistent knowledge base for multi-turn AI agent interactions
• Implement context augmentation by searching stored user preferences or prior interactions before generating responses
• Build domain-specific memory for specialized AI workflows (e.g., customer service agents recalling past tickets)

Setup pitfalls

• Cannot set both QDRANT_URL and QDRANT_LOCAL_PATH simultaneously; choose one deployment mode
• Remote Qdrant requires QDRANT_API_KEY; omitting it silently fails with connection errors
• Embedding model choice affects search quality and memory footprint; all-MiniLM-L6-v2 is lightweight but less capable for specialized domains
• FastMCP server binds to 127.0.0.1:8000 by default; networking may fail if your MCP client expects a different host or port