---
title: "Qdrant MCP Server: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/qdrant/
description: "Stores text plus metadata and retrieves semantic matches from a configured Qdrant collection; its outputs are context text, not a scored retrieval-evaluation dataset."
---

# Qdrant MCP Server: review, setup & limitations · Undominated.ai

> Stores text plus metadata and retrieves semantic matches from a configured Qdrant collection; its outputs are context text, not a scored retrieval-evaluation dataset.

[← Explore all mcp servers](/mcp-servers/)

VECTOR MEMORY / Qdrant

# Qdrant MCP Server

Stores text plus metadata and retrieves semantic matches from a configured Qdrant collection; its outputs are context text, not a scored retrieval-evaluation dataset.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/qdrant/mcp-server-qdrant)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 7 linked sources
 Publisher Qdrant
 Licence [Apache-2.0 ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/LICENSE)
 Revision c56ae5adf62b
 [Read what was—and wasn’t—checked ↓](#review)

“An official Model Context Protocol server for keeping and retrieving memories in the Qdrant vector search engine.”

 [Qdrant · upstream description ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/README.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The actual tool registration removes qdrant-store when QDRANT_READ_ONLY is enabled, allowing a narrower retrieval setup.
- The connector exposes how information and metadata become payloads, and the settings define collection binding, result count and optional filter exposure.

### A good fit for

 - Retrieving approved notes or snippets from a known collection using a natural-language query and a compatible embedding model.
- Preparing retrieved context for a downstream answer/evaluation workflow when source identity and provenance are stored explicitly in metadata.

### Weigh up before choosing

 - Storage is enabled by default. QDRANT_READ_ONLY removes the storage tool but does not sandbox embedding-model downloads, local cache access or the client's database credentials.
- The connector drops Qdrant point IDs and similarity scores before formatting results. The tool provides no pagination, score threshold, stable citation ID or retrieval-quality measurement.
- A missing collection and a valid query with no matches both become no results. Do not treat that output as proof an indexed corpus was searched successfully.
- Each store call generates a new UUID and returns a text confirmation without the point ID; retrying the same content can create duplicates. There is no delete/dedup/update tool in this reviewed surface.
- New collections and configured payload indexes are created on the store path, not by find. Existing collections must already match the embedding vector name/dimension; the connector does not migrate them.
- Only FastEmbed is implemented. Model weights/licences, FastEmbed, qdrant-client, FastMCP and transitive dependencies were not audited. The uvx command does not pin the reviewed source or resolved package graph.
- The README describes loopback HTTP/SSE defaults but no listener authentication. Returned document text is interpolated into entry tags without escaping and must remain untrusted data, not instructions.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - README.md, pyproject.toml and LICENSE
- src/mcp_server_qdrant/server.py
- src/mcp_server_qdrant/mcp_server.py
- src/mcp_server_qdrant/settings.py
- src/mcp_server_qdrant/qdrant.py
- src/mcp_server_qdrant/embeddings/fastembed.py and factory.py
- src/mcp_server_qdrant/common/wrap_filters.py

### Our findings

 - Store accepts information and optional object metadata; find accepts query. When COLLECTION_NAME is set, collection_name is bound out of both exposed signatures; otherwise the caller supplies it. Optional filter arguments depend on filterable_fields or QDRANT_ALLOW_ARBITRARY_FILTER.
- find returns a query header and entry-tagged strings with content and JSON metadata, or None. Its connector constructs Entry from payload document/metadata only, discarding score and point ID; the default result limit is 10.
- store embeds the text, creates a missing collection with cosine vectors and optional indexes, and upserts a fresh random ID. A workflow needing reproducible retrieval metrics needs a separate measurement/identity path.
- settings.py rejects local_path with either location or api_key. FastEmbed uses distinct passage_embed/query_embed calls and names the vector from the model's final path segment.
- The manifest declares Python >=3.10, FastMCP ==2.7.0, FastEmbed >=0.6.0, qdrant-client >=1.12.0 and pydantic >=2.10.6,<2.12.0, built with hatchling. These are inspected constraints, not installed versions or advisory clearance.
- This narrows the prior generic automatic-collection wording to the actual store path. Existing config and install receipts remain unchanged.

### Not established by this review

 - No installation, model download, server startup, MCP handshake, Qdrant connection, read, write or filter call was performed.
- No embedding compatibility, retrieval relevance, duplicate handling, listener authentication or package-advisory audit. Remaining filter/partial-function helpers and third-party implementations were not exhaustively audited.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/README.md)

DOCUMENTED COMMAND

 uvx mcp-server-qdrant Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - Set the Qdrant URL or local path and collection configuration.
- For retrieval-only access, set QDRANT_READ_ONLY=true and use appropriately scoped Qdrant credentials.

### Before you start

 - uv/Python 3.10 or newer, a Qdrant endpoint or local data path, and storage for the embedding model.

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "qdrant": {
 "command": "uvx",
 "args": [
 "mcp-server-qdrant"
 ],
 "env": {
 "QDRANT_URL": "https://xyz-example.eu-central.aws.cloud.qdrant.io:6333",
 "QDRANT_API_KEY": "your_api_key",
 "COLLECTION_NAME": "your-collection-name",
 "EMBEDDING_MODEL": "sentence-transformers/all-MiniLM-L6-v2"
 }
 }
} Copy configuration ↗

### Compatibility

An MCP client supporting stdio, SSE, Streamable HTTP. · uv/Python, a Qdrant endpoint or local data path, and storage for the embedding model.

### Implementation

Python

### Transports

stdio · SSE · Streamable HTTP

### Vector memory / Configured credentials or local access

 - Read text/metadata matches from the configured collection or, when unbound, caller-selected collections available to the credentials.
- Unless disabled, store writes embeddings and metadata and can create collections/indexes.
- Open QDRANT_LOCAL_PATH or authenticate to QDRANT_URL with QDRANT_API_KEY; embedding setup uses local compute/cache and may download weights. Debug messages include query/stored text.

### Cost model

Qdrant hosting and local embedding compute determine cost.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Pinned setup and original configuration ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/README.md) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/README.md Supports: upstreamDescription, install, compatibility, transports, configExample, limitations
- [Actual tool signatures, registration and output formatting ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/mcp_server.py) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/mcp_server.py Supports: summary, whySelected, bestFor, review, limitations, access
- [Storage/search payload and collection behavior ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/qdrant.py) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/qdrant.py Supports: review, limitations, access
- [Configuration defaults and constraints ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/settings.py) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/settings.py Supports: review, limitations
- [Embedding adapter ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/embeddings/fastembed.py) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/src/mcp_server_qdrant/embeddings/fastembed.py Supports: review, limitations
- [Dependency/build manifest ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/pyproject.toml) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/pyproject.toml Supports: review, limitations
- [Repository licence ↗](https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/LICENSE) Checked 2026-10-14 https://github.com/qdrant/mcp-server-qdrant/blob/c56ae5adf62bb78d852bf7bbcbc5d7b75e2bbe41/LICENSE Supports: license

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### AWS IaC MCP Server ↗ Validates CloudFormation templates, retrieves CDK/CloudFormation guidance and investigates failed deployments.](/mcp-servers/aws-iac/)[### AWS Labs OpenAPI MCP ↗ Pins the OpenAPI MCP server that builds tools from an operator-supplied spec and names stdio as the transport.](/mcp-servers/awslabs-openapi/)[### Chroma MCP ↗ Lets an assistant create and query Chroma collections through local, HTTP or Chroma Cloud database connections.](/mcp-servers/chroma/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
