Qdrant MCP Server
Officialby Qdrant
The Qdrant MCP Server is an official Model Context Protocol server built and maintained by the Qdrant team that turns Qdrant's high-performance vector search engine into a semantic memory layer for LLM applications. It exposes two simple MCP tools — qdrant-store and qdrant-find — that allow any MCP-compatible AI client (Claude Desktop, Cursor, Windsurf, VS Code, Claude Code) to persistently store and semantically retrieve information at runtime. Under the hood it uses FastEmbed to generate vector embeddings locally, so no external embedding API key is required. Collections are created automatically on first use, and the server supports both a remote Qdrant instance via QDRANT_URL and an embedded local database via QDRANT_LOCAL_PATH. It ships with three transport protocols — stdio, SSE, and Streamable HTTP — making it equally at home as a local tool or a shared remote service. Tool descriptions are fully overridable via environment variables, enabling specialized deployments such as a semantic code-snippet store for Cursor or a project knowledge base for Claude Code.
Installation
Key Features
- ✓Persistent semantic memory: store and retrieve information from Qdrant using natural-language queries, not exact-match keywords
- ✓Zero-config embeddings via FastEmbed (sentence-transformers/all-MiniLM-L6-v2) — no external embedding API needed
- ✓Supports remote Qdrant Cloud, self-hosted Qdrant, and fully local embedded mode with no server required
- ✓Three transport protocols: stdio (local), SSE, and Streamable HTTP for remote and team deployments
- ✓Customizable tool descriptions via environment variables — repurpose as a code-snippet store or project knowledge base
- ✓Auto-creates Qdrant collections on first use with any MCP-compatible client
Use Cases
- →Give Claude Desktop or Claude Code a long-term semantic memory that persists across conversations
- →Build a semantic code-snippet library inside Cursor or Windsurf — store snippets with natural-language descriptions
- →Index project documentation or team knowledge so AI agents can look up relevant context on demand
- →Power RAG pipelines by letting the AI store and search its own context window overflow
- →Share a single remote MCP server across an entire engineering team for a shared knowledge store