Skip to directory

About Qdrant MCP

Qdrant's official local MCP server stores and retrieves text memories using semantic search backed by a Qdrant database or local Qdrant storage.

What you can use it for

  • Save project notes and retrieve them semantically later
  • Build a lightweight persistent memory layer over Qdrant
  • Store and find code snippets by configuring tool descriptions
Connection and setup
Hosting
Local stdio by default; optional local SSE or Streamable HTTP server
Transport
stdio, SSE, Streamable HTTP
Endpoint
Local process; no hosted endpoint documented
Authentication
No MCP authentication is documented for local stdio. Qdrant API key is optional when connecting to a secured Qdrant endpoint; local database mode uses QDRANT_LOCAL_PATH.

Requirements

  • Python package runner uvx or Docker
  • A Qdrant server URL or local Qdrant database path (not both)
  • Collection name; server creates it if absent
  • FastEmbed embedding provider; documented default model sentence-transformers/all-MiniLM-L6-v2

Setup steps

  1. Configure QDRANT_URL and optionally QDRANT_API_KEY for Qdrant Cloud/self-hosted service, or set QDRANT_LOCAL_PATH for local storage; these two modes cannot be set together.
  2. Set COLLECTION_NAME and optionally EMBEDDING_MODEL.
  3. Add local stdio configuration using command uvx and args ["mcp-server-qdrant"], with the required environment values. Start with a test collection or local path.
  4. For a remote client, start with --transport streamable-http or --transport sse and configure the client to the resulting local/network endpoint; protect any exposed HTTP service with network controls and authentication at the proxy.
  5. Verify with qdrant-find on a harmless query. Use qdrant-store only when you intend to write information.
Tools 2

The official Qdrant reference lists two tools. QDRANT_READ_ONLY=true removes qdrant-store, so the configured runtime can expose one or two.

  • qdrant-store

    Store information and optional JSON metadata in a Qdrant collection; requires collection_name if no default collection is configured.

    01
  • qdrant-find

    Search for relevant stored information using a text query and return matching messages.

    02
Limitations and notes
  • One configured default collection name is intended per server process; use explicit collection names when needed.
  • The documented embedding provider is FastEmbed; vectorization requires downloading/using the selected embedding model.
  • SSE/Streamable HTTP deployment is self-hosting, not a Qdrant-provided hosted endpoint. Secure it before exposing beyond a trusted network.
  • Store is write-capable and can create the collection automatically; validate target before enabling writes.
Original sources
  • official tool reference

    Official docs list qdrant-store and qdrant-find. QDRANT_READ_ONLY removes the store tool, so live availability is configuration-dependent.

  • official documentation or source

    Official docs list qdrant-store and qdrant-find. QDRANT_READ_ONLY removes the store tool, so live availability is configuration-dependent.

Documentation checked 2026-10-28T12:53:18.000Z. Connection has not been runtime tested.

Similar tools