Qdrant is an MCP server that provides a semantic memory layer connected directly to the Qdrant vector database. Built using FastMCP, it connects LLM applications such as Claude Desktop, Cursor, and Windsurf to remote or local Qdrant vector storage instances. Software engineers and AI developers use this server to equip language models with long-term memory, context persistence, and vector-based document or code snippet retrieval. The server exposes two primary tools: qdrant-store for vectorizing and indexing text payloads alongside structured metadata, and qdrant-find for executing natural language semantic searches across stored records. Embeddings are generated locally using the FastEmbed library, defaulting to sentence-transformers/all-MiniLM-L6-v2 without requiring external embedding APIs. It supports multiple transport protocols including standard I/O (stdio), Server-Sent Events (SSE), and streamable HTTP. By configuring custom tool prompts and read-only modes, users can restrict the assistant to query-only retrieval or turn the database into a specialized semantic code repository.
Category: AI Memory & Context
Tags: embeddings, qdrant, rag, semantic search, vector-database
uvx installed and have access to a running Qdrant instance or local database directory. 2. For automatic installation in Claude Desktop, run: npx @smithery/cli install mcp-server-qdrant --client claude. 3. Alternatively, open your claude_desktop_config.json and add the server to the mcpServers object: json { "mcpServers": { "qdrant": { "command": "uvx", "args": ["mcp-server-qdrant"], "env": { "QDRANT_URL": "http://localhost:6333", "COLLECTION_NAME": "my-collection", "EMBEDDING_MODEL": "sentence-transformers/all-MiniLM-L6-v2" } } } } 4. For Cursor or Windsurf, launch the server over SSE by running uvx mcp-server-qdrant --transport sse and connect the client to http://localhost:8000/sse.qdrant-store tool for persistent agent memory. - Querying relevant documentation, notes, or previous conversations via semantic similarity searches using the qdrant-find tool. - Saving reusable code snippets with metadata in Cursor or Windsurf to retrieve relevant programming patterns based on natural language descriptions. - Operating a shared team knowledge base in read-only mode by enabling QDRANT_READ_ONLY to restrict agents to search operations.Part of MCP Servers
You can install it automatically using the Smithery CLI with npx @smithery/cli install mcp-server-qdrant --client claude. You can also run it directly using uvx mcp-server-qdrant by configuring your claude_desktop_config.json or running a pre-built Docker container.
It serves as a semantic memory layer on top of the Qdrant vector database. The server enables LLMs to persist arbitrary text and metadata into vector collections with qdrant-store, and retrieve contextually relevant entries using natural language semantic queries with qdrant-find.
It functions with any MCP-compliant client. Supported clients explicitly tested include Claude Desktop via stdio transport, as well as AI coding environments like Cursor and Windsurf via Server-Sent Events (SSE) or streamable HTTP transports.
The server currently uses FastEmbed to generate text embeddings locally. Its default model is sentence-transformers/all-MiniLM-L6-v2, which can be modified by setting the EMBEDDING_MODEL environment variable during server initialization.
Yes, it is open source and hosted on GitHub by the official Qdrant team. Developers can inspect, fork, or build the codebase directly from source or run it using Docker.