Qdrant MCP server
Qdrant's official MCP server: a semantic memory layer that stores and retrieves information from a Qdrant vector database.
The Qdrant MCP server is the official reference implementation that turns a Qdrant vector database into a semantic memory layer for AI agents. It exposes two deliberately small tools: one to store a piece of information (with optional structured metadata) and one to retrieve the most relevant stored information for a natural-language query. The server handles embedding generation internally using FastEmbed by default, so an agent can persist notes, code snippets, decisions, or documents and later recall them by meaning rather than exact keywords, making it a drop-in long-term memory or knowledge-retrieval backend.
The server is published as the mcp-server-qdrant Python package and is typically launched with uvx (uvx mcp-server-qdrant). It connects to a Qdrant instance via QDRANT_URL (or a local on-disk path through QDRANT_LOCAL_PATH) and an optional QDRANT_API_KEY for hosted clusters, writing to the collection named by COLLECTION_NAME. The embedding model is configurable through EMBEDDING_MODEL, and the two tool descriptions can be customized so the same server reads as a memory tool, a code-search tool, or a documentation index depending on how you frame it for the agent. A read-only mode and a search-result limit round out the configuration.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"qdrant": {
"command": "uvx",
"args": [
"mcp-server-qdrant"
],
"env": {
"QDRANT_URL": "<QDRANT_URL>",
"COLLECTION_NAME": "<COLLECTION_NAME>",
"QDRANT_API_KEY": "<QDRANT_API_KEY>",
"EMBEDDING_MODEL": "<EMBEDDING_MODEL>"
}
}
}
}claude mcp add qdrant -- uvx mcp-server-qdrantAvailable tools
| Tool | Description |
|---|---|
| qdrant-store | Stores a piece of information, with optional metadata, in the Qdrant collection. |
| qdrant-find | Retrieves the most relevant stored information from the Qdrant collection for a natural-language query. |
Required configuration
- QDRANT_URLRequired
URL of the Qdrant instance. Required unless using QDRANT_LOCAL_PATH.
- COLLECTION_NAMERequired
Name of the Qdrant collection to use. Required unless provided per-tool.
- QDRANT_API_KEYOptional
API key for a remote/hosted Qdrant cluster. Optional.
- QDRANT_LOCAL_PATHOptional
Path to a local on-disk Qdrant database. Optional; mutually exclusive with QDRANT_URL.
- EMBEDDING_MODELOptional
Embedding model used to encode text. Optional; defaults to sentence-transformers/all-MiniLM-L6-v2.
- QDRANT_READ_ONLYOptional
Disables write operations when set. Optional; defaults to false.
What you can do with it
Long-term memory for an agent
Persist facts, decisions, and snippets with qdrant-store as the agent works, then recall the relevant ones later with qdrant-find by semantic similarity, giving the agent continuity across sessions.
Semantic code or document search
Index a codebase or knowledge base into a collection and let the agent retrieve the most relevant passages for a question, customizing the tool descriptions so the agent frames it as code search or doc lookup.
FAQ
- Is it free?
- Yes. The MCP server is open source and free to run, and Qdrant itself is open source and self-hostable at no cost. You only pay if you choose Qdrant Cloud for a managed cluster.
- Does it support remote/OAuth?
- It runs locally over stdio (via uvx) and connects to your Qdrant instance with QDRANT_URL and an optional QDRANT_API_KEY. There is no OAuth flow; authentication is handled by Qdrant's own API key, and the Qdrant instance itself can be local or a remote managed cluster.