Qdrant MCP server
Vector similarity engine for semantic embeddings, hybrid search, and persistent AI agent memory.
Local (stdio)Credential
Set up Qdrant Vector DB
Configuration examples, not tested connections. Replace placeholders and review the upstream documentation before use.
Claude Code
Run in your terminal.
claude mcp add qdrant --env QDRANT_URL="http://localhost:6333" --env COLLECTION_NAME="$COLLECTION_NAME" -- uvx mcp-server-qdrant
Cursor
Merge into ~/.cursor/mcp.json
{
"mcpServers": {
"qdrant": {
"command": "uvx",
"args": [
"mcp-server-qdrant"
],
"env": {
"QDRANT_URL": "http://localhost:6333",
"COLLECTION_NAME": "REPLACE_WITH_COLLECTION_NAME"
}
}
}
}
Codex
Run in your terminal.
codex mcp add qdrant --env QDRANT_URL="http://localhost:6333" --env COLLECTION_NAME="$COLLECTION_NAME" -- uvx mcp-server-qdrant
Gemini CLI
Merge into ~/.gemini/settings.json
{
"mcpServers": {
"qdrant": {
"command": "uvx",
"args": [
"mcp-server-qdrant"
],
"env": {
"QDRANT_URL": "http://localhost:6333",
"COLLECTION_NAME": "REPLACE_WITH_COLLECTION_NAME"
}
}
}
}
OpenCode
Merge into ~/.config/opencode/opencode.jsonc
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"qdrant": {
"type": "local",
"command": [
"uvx",
"mcp-server-qdrant"
],
"enabled": true,
"environment": {
"QDRANT_URL": "http://localhost:6333",
"COLLECTION_NAME": "REPLACE_WITH_COLLECTION_NAME"
}
}
}
}
VS Code
Merge into .vscode/mcp.json
{
"servers": {
"qdrant": {
"type": "stdio",
"command": "uvx",
"args": [
"mcp-server-qdrant"
],
"env": {
"QDRANT_URL": "http://localhost:6333",
"COLLECTION_NAME": "${input:COLLECTION_NAME}"
}
}
},
"inputs": [
{
"id": "COLLECTION_NAME",
"type": "promptString",
"description": "COLLECTION_NAME",
"password": false
}
]
}
Using more than one server?Add to a config with other servers
Step-by-step setup guides: Claude Code and Codex CLI, Cursor and VS Code. In Claude Desktop, the Cursor JSON goes in claude_desktop_config.json. This is a local server that runs on your machine; see remote vs local servers and authentication and security basics.
Overview
mcp-server-qdrant is the official Model Context Protocol server from Qdrant, the vector search engine. Upstream describes it as a semantic memory layer: your agent can save pieces of information to a Qdrant collection and later find them again by meaning rather than exact wording.
It has two tools. qdrant-store saves text, with optional JSON metadata, into a collection; qdrant-find runs a semantic search and returns matching entries. Embeddings are created locally with FastEmbed, using sentence-transformers/all-MiniLM-L6-v2 by default. Use it to give an assistant long-term memory across sessions, or, as upstream shows, as a code-search tool for Cursor, Windsurf or Claude Code by overriding the tool descriptions so the agent stores and retrieves code snippets.
Setup: the server runs with uvx mcp-server-qdrant, so you need uv installed, and it is configured entirely through environment variables. Set QDRANT_URL for a running Qdrant instance, or QDRANT_LOCAL_PATH for an on-disk database, but not both. Add QDRANT_API_KEY if your Qdrant server requires an API key. COLLECTION_NAME sets the default collection; without it, every tool call must name one. QDRANT_READ_ONLY=true removes the store tool, which is useful when an agent should only search. The snippets below assume a local Qdrant on port 6333. stdio is the default transport; SSE and Streamable HTTP are available with --transport if you want to run it as a shared server.
Example tools
qdrant-storeqdrant-find
Tool names are upstream examples. The server's live tool list can differ.
Related Database servers
Sources
This page was written from the upstream documentation below and last reviewed on 11 October 2026. Listing is not a security or compatibility endorsement. Report a correction through the contact page.