Verified

Qdrant MCP Server

Vector search capabilities for Qdrant databases within LLM-powered applications.

Data last scanned today · Reviewed today

582 tools★ 1,481PythonApache-2.0

Overview

This server provides a semantic memory layer for LLM applications by connecting them to a Qdrant vector database. It allows AI assistants to store and retrieve contextual information directly within the database infrastructure. The server exposes two primary tools: qdrant-store for saving information with optional metadata, and qdrant-find for executing semantic searches. Configuration is managed through environment variables, supporting both remote Qdrant server URLs and local file paths.

Our verdict

With a security score of 53/100, this server is currently classified as high-risk. This rating is largely due to the fact that it is a community-maintained project rather than an official Qdrant vendor implementation, and key security aspects like authentication methods and read-only mode support have not yet undergone editorial review. It is suitable for experimental workflows or internal tools, but users should carefully evaluate its integration in sensitive environments.

Zero direct dependencies reduce the supply-chain attack surface.
Active development evidenced by recent commits.
Supports both remote server URLs and local database paths.

Security architecture has not yet undergone editorial review.
Community-maintained without official vendor oversight.
Authentication and read-only modes remain unverified.

Tools

Tool	Description	Risk
qdrant-store	Store some information in the Qdrant database	medium
qdrant-find	Retrieve relevant information from the Qdrant database	low

Compatibility

Client	Local	Docker	Remote	Read-only
Claude Desktop
Cursor
VS Code
Windsurf
ChatGPT

Frequently asked questions

›Can I connect to a local Qdrant database instead of a remote server?

Yes, you can provide a local path using the QDRANT_LOCAL_PATH environment variable. Note that you cannot specify both QDRANT_URL and QDRANT_LOCAL_PATH simultaneously.

›How can I prevent the server from writing data to my database?

You can enable read-only mode by setting the QDRANT_READ_ONLY environment variable to true, which disables the qdrant-store tool.

›Is it possible to override the default embedding model?

Yes, you can specify a different model using the EMBEDDING_MODEL environment variable, though currently only the fastembed provider is supported.

›Does the server require a default collection name?

A default collection name is not strictly required, but if you do not set one, you must provide the collection_name parameter every time you call the qdrant-store or qdrant-find tools.

›How many search results does the server return by default?

The server returns a maximum of 10 search results by default, though this limit can be adjusted using the QDRANT_SEARCH_LIMIT environment variable.