mcp-server-qdrant
An MCP server that stores and retrieves memories in the Qdrant vector search engine, acting as a semantic memory layer.
Tested on LinuxKnown vulnerabilities
Use it
Before you install: known vulnerabilities
pickfit found 5 known high or critical vulnerabilities in the packages a fresh install brings (OSV.dev, 16 Nov 2026). Weigh this before you install it.
pypi:fastmcp 2.7.0FastMCP OpenAPI Provider has an SSRF & Path Traversal Vulnerability (GHSA-vv7q-7jx5-f767, critical, fixed in 3.2.0)
See 4 more
pypi:fastmcp 2.7.0— GHSA-5h2m-4q8j-pqpj (high), fixed in 2.14.2pypi:fastmcp 2.7.0— GHSA-c2jp-c369-7pvx (high), fixed in 2.13.0pypi:fastmcp 2.7.0— GHSA-rcfx-77hg-w2wv (high), fixed in 2.14.0pypi:fastmcp 2.7.0— GHSA-rww4-4w9c-7733 (high), fixed in 3.2.0
Linux x86_64
uv tool install mcp-server-qdrant and pre-download the default FastEmbed model so the server works offline
uv tool install mcp-server-qdrant && ~/.local/share/uv/tools/mcp-server-qdrant/bin/python -c "from fastembed import TextEmbedding; m=TextEmbedding('sentence-transformers/all-MiniLM-L6-v2'); print(len(list(m.embed(['hi']))[0]))"
Add it to your agent
claude mcp add mcp-server-qdrant -- mcp-server-qdrantJSON for Claude Desktop, Cursor and other clients
{
"mcpServers": {
"mcp-server-qdrant": {
"command": "mcp-server-qdrant",
"args": []
}
}
}Add mcp-server-qdrant as a server in your MCP client, setting QDRANT_LOCAL_PATH (or a Qdrant URL) and COLLECTION_NAME in its environment, and the client will list the qdrant-store and qdrant-find to
What it does
- Store information in a Qdrant collectionTested
How to use it
- Start
mcp-server-qdrantfrom your MCP client with QDRANT_LOCAL_PATH=<folder> and COLLECTION_NAME=<collection> set in its environment. - Call the qdrant-store tool with an
informationtext and optionalmetadata.
- Start
- Find stored information by meaningTested
How to use it
- Store some information with the qdrant-store tool.
- Call the qdrant-find tool with a
querywritten in plain language, such as <your question>.
- Search code snippets by meaning for coding agentsTested
How to use it
- Start
mcp-server-qdrantwith COLLECTION_NAME=<collection>, and set TOOL_STORE_DESCRIPTION and TOOL_FIND_DESCRIPTION to describe storing and finding code snippets. - Call qdrant-store with a description in
informationand the code inmetadata, such as {"code": "<your code>"}. - Call qdrant-find with a
querydescribing the code you need.
- Start
How far pickfit got
Linux x86_64
- Installs
- Connects
- 3 of 3 features work
- Works offline
Good for
- store code snippets for later retrieval
- search code examples semantically
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
An MCP client called one of its tools
- Replayed from scratch
- Works offline
- Checked on 15 Nov 2026
- Version 0.8.1
An MCP client handshake over stdio listed 2 tools, and a call to qdrant-store returned a result.
Differences the test machine hit
- README: Examples use QDRANT_URL=http://localhost:6333 (a running Qdrant server)
In the test: Used QDRANT_LOCAL_PATH=/tmp/qdrant-db (embedded local mode, documented in the README) because no Qdrant server runs here - README: uvx mcp-server-qdrant needs no specific installation
In the test: On first use the server downloads the FastEmbed model sentence-transformers/all-MiniLM-L6-v2 from Hugging Face. To make it work offline, the model was downloaded ahead of time in the install step. - README: Code search example uses --transport sse for Cursor/Windsurf
In the test: Tested the same store/find with custom tool descriptions over stdio