Qdrant MCP Server
mcp-server-qdrant
Qdrant's official MCP server: semantic memory for agents on a vector database, storing and retrieving by meaning
Install
QDRANT_URL="http://localhost:6333" \
COLLECTION_NAME="my-collection" \
EMBEDDING_MODEL="sentence-transformers/all-MiniLM-L6-v2" \
uvx mcp-server-qdrantRuns over stdio; add --transport sse or streamable-http for remote clients.
This is third-party code. Review the repository files before installing.
What it does
The server turns Qdrant into long-term agent memory with two tools. qdrant-store saves text with optional metadata into a collection, and qdrant-find retrieves stored items by query meaning. Embeddings are computed locally with FastEmbed models, all-MiniLM-L6-v2 by default. It can connect to a Qdrant server by URL or to a local database in a folder. Tool descriptions are configurable, so the server can be repurposed, for example as a code snippet store.
Who it is for. For developers who need shared agent memory across sessions and clients on their own vector database.
Good fit when
- The agent must remember decisions and facts across sessions
- You need semantic search over accumulated notes or code snippets
- Qdrant is already used in the project
Not a fit when
- A plain memory file like CLAUDE.md is enough
- You need embeddings from an external provider: only FastEmbed is supported
Example request
Remember that this project stores dates in UTC, then find all stored agreements about data formatsLimitations
Only FastEmbed models are supported. QDRANT_URL and QDRANT_LOCAL_PATH cannot be set together. The embedding model is downloaded on first run. Local mode and self-hosted Qdrant need no external services; Qdrant Cloud needs a key.
How to disable. Remove the qdrant block from the MCP client config. Stored data stays in the Qdrant collection or local folder and is deleted separately.
MCP
- Transport
- stdio, sse, http
- Authentication
- API key
| Environment variables | |
|---|---|
| QDRANT_URL | Qdrant server URL, alternative to QDRANT_LOCAL_PATH |
| QDRANT_API_KEY secret | Qdrant server API key |
| QDRANT_LOCAL_PATH | Path to a local Qdrant database |
| COLLECTION_NAME | Default collection |
| EMBEDDING_MODEL | FastEmbed embedding model |
| QDRANT_READ_ONLY | true disables writes |
Security check
- Writes data to a vector database
- Downloads an embedding model on first run
README in short
The README describes the two tools, all server and FastMCP environment variables, and running via uvx, Docker and Smithery. It covers stdio, SSE and Streamable HTTP transports and Claude Desktop configs for cloud and local Qdrant. There are setup examples for Cursor, Windsurf and VS Code, including use as a code store. Apache 2.0 licensed.
FAQ
Do I need a running Qdrant?
No, set QDRANT_LOCAL_PATH and the database is stored in a local folder.
Can I block writes?
Yes, QDRANT_READ_ONLY=true disables the qdrant-store tool.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent