Qdrant Vector Memory

degraded · temporarily disabled

qdrant

Store and retrieve semantic memories in the Qdrant vector search engine via the official Qdrant MCP server.

Installs: —
Namespace: qdrant
Version: >=0.8.0 <1.0.0
Last attach: reviewed1 months ago

Add to your mcp.json

{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "<QDRANT_URL>",
        "QDRANT_API_KEY": "<QDRANT_API_KEY>"
      }
    }
  }
}

Open in VS CodeDon't have it? Install FrootAI for VS Code

Attach in a FrootAI play (fai-manifest)

mcp_scope:
  attached:
    - qdrant

Authentication

api-key

create a Qdrant Cloud cluster (cloud.qdrant.io) and set QDRANT_URL + QDRANT_API_KEY; or run a local Qdrant and point QDRANT_URL at http://localhost:6333.

Required credentials: QDRANT_URLQDRANT_API_KEY

Environment variables

Variable	Required	Auth mode	Description
QDRANT_URL	yes	any	URL of the Qdrant server (e.g. https://xyz.cloud.qdrant.io:6333 for Qdrant Cloud, or http://localhost:6333 for a local instance). An embedded QDRANT_LOCAL_PATH may be used instead.
QDRANT_API_KEY	yes	api-key	API key for the Qdrant server. Required for Qdrant Cloud; optional for a local/unauthenticated instance. Pass via env, never inline in args (doctrine #6).
COLLECTION_NAME	no	any	Name of the default collection to read/write. Created automatically if it does not exist.
EMBEDDING_MODEL	no	any	FastEmbed model used to encode memories (default sentence-transformers/all-MiniLM-L6-v2). Only FastEmbed models are supported.
QDRANT_READ_ONLY	no	any	When set to true, disables the qdrant-store write tool — a read-only retrieval posture.

Sample tools

qdrant.qdrant-store
Store information (with optional metadata) as a memory in a Qdrant collection.
qdrant.qdrant-find
Retrieve relevant memories from a Qdrant collection via semantic search.

Used in recipes

FAI cookbook recipes that attach qdrant.

Destructive-action audit

No destructive tools — audited read-only

Destructive-action audit (X2.18): no destructive tools — qdrant-store appends a memory and qdrant-find reads; neither drops collections or removes points.

Known limitations

First run downloads the FastEmbed embedding model (default sentence-transformers/all-MiniLM-L6-v2) into the local cache, adding startup latency and disk usage; only FastEmbed models are supported.
Runs via `uvx mcp-server-qdrant`, so the host needs uv / Python available on PATH; the package is fetched and installed on first launch.
Targeting Qdrant Cloud requires QDRANT_URL plus a QDRANT_API_KEY credential; a local instance or embedded QDRANT_LOCAL_PATH can run unauthenticated, and QDRANT_READ_ONLY=true disables the write tool.

Spec contributors

Maintainers who landed commits onmcp-specs/qdrant.json.

pspsbali

Freshness

When this spec was last reviewed by a maintainer and last machine-validated against the live server.

Spec last reviewed: 1 months ago(2026-06-25)
Auto-validated: never