# Qdrant

> Open-source vector search with payload filtering, self-hostable or managed.

[Canonical HTML page](https://top-mcps.com/mcp/qdrant) · [server.json](https://top-mcps.com/mcp/qdrant.json) · [methodology](https://top-mcps.com/about/methodology)

## Install

### Claude Desktop — `claude_desktop_config.json`

Paste under mcpServers. Fully quit and reopen Claude after editing.

```json
{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### Claude Code — `CLI or .mcp.json`

Run from your repo. Commit .mcp.json to share with your team.

```shell
# export QDRANT_URL=https://your-cluster.qdrant.io
# export QDRANT_API_KEY=YOUR_API_KEY
claude mcp add qdrant -- uvx mcp-server-qdrant
```

### Cursor — `.cursor/mcp.json`

Global path: ~/.cursor/mcp.json. Reload window after editing.

```json
{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### VS Code — `.vscode/mcp.json`

VS Code uses the "servers" key (not "mcpServers").

```jsonc
{
  "servers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### Windsurf — `~/.codeium/windsurf/mcp_config.json`

Open via Cascade → hammer icon → Configure.

```json
{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### Cline — `cline_mcp_settings.json`

Open via the Cline sidebar → MCP Servers → Edit.

```json
{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": [
        "mcp-server-qdrant"
      ],
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### Continue — `~/.continue/config.json`

Continue uses modelContextProtocolServers with a transport block.

```json
{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "uvx",
          "args": [
            "mcp-server-qdrant"
          ],
          "env": {
            "QDRANT_URL": "${QDRANT_URL}",
            "QDRANT_API_KEY": "${QDRANT_API_KEY}"
          }
        }
      }
    ]
  }
}
```

### Codex CLI — `~/.codex/config.toml`

Codex uses TOML. Each server is a [mcp_servers.<name>] subtable.

```shell
# ~/.codex/config.toml
[mcp_servers.qdrant]
command = "uvx"
args = [
  "mcp-server-qdrant",
]
env = { QDRANT_URL = "${QDRANT_URL}", QDRANT_API_KEY = "${QDRANT_API_KEY}" }
```

### Zed — `~/.config/zed/settings.json`

Zed calls them "context_servers". Settings live-reload on save.

```jsonc
{
  "context_servers": {
    "qdrant": {
      "command": {
        "path": "uvx",
        "args": [
          "mcp-server-qdrant"
        ]
      },
      "env": {
        "QDRANT_URL": "${QDRANT_URL}",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}"
      }
    }
  }
}
```

### ChatGPT — `ChatGPT → Apps directory`

Qdrant doesn't ship a hosted HTTPS endpoint today. ChatGPT supports remote MCP servers only — to use this server in ChatGPT you'll need to deploy it to a public HTTPS URL first (e.g. via Cloudflare Workers or Vercel) or wait for an official remote build.

```none

```

## At a glance

- **Maintainer:** Qdrant
- **Transport:** stdio
- **Auth model:** API key
- **Required secrets:** QDRANT_URL, QDRANT_API_KEY
- **Supported clients:** Claude, Cursor, Any MCP-compatible client
- **License:** Apache-2.0
- **Language:** Python
- **Latest version:** latest
- **Last verified:** 2026-05-26
- **GitHub stars:** 1,410 (fetched 2026-05-26T10:50:23.189Z)
- **Score:** 84/100 (rubric 2026-04 — see https://top-mcps.com/about/methodology)
- **Source:** https://github.com/qdrant/mcp-server-qdrant

## Security & scope

- **Access scope:** read-write
- **Sandbox:** Local mode has no auth by default — protect via network policy. Cloud mode uses API keys.
- **Gotchas:**
  - Local default port (6333) is bound to 0.0.0.0 in some Docker setups — bind to localhost explicitly.
  - No multi-tenant primitives in OSS — separate by collection.

## Quick answer

**What it does.** Provides MCP tools for collection creation, point upserts, similarity queries with payload filters, batch operations, and snapshot management.

**Best for:**
- Self-hosted vector search
- Payload-filtered retrieval
- Local development with cloud parity
- Open-source compliance requirements

**Not for:**
- Hands-off ops
- Workloads relying on proprietary vendor features

## Description

The Qdrant MCP exposes Qdrant's collection management and search primitives. Open-source under Apache 2.0, runs locally via Docker or as a managed service, with strong support for filtering on metadata payloads alongside vector similarity.

## Why it matters

Vendor lock-in worries are real for retrieval stacks. Qdrant's license, self-host story, and managed offering give an exit ramp the cloud-only vendors do not. The MCP version makes wiring it to an agent trivial.

## Key features

- Apache-2.0 license
- Strong payload filters
- Self-host or managed
- Hybrid sparse-dense
- Quantization for memory

## FAQ

### How does the self-host story really work?

Docker one-liner: `docker run -p 6333:6333 qdrant/qdrant`. Point the MCP at `http://localhost:6333` with no API key. Production self-host is a Kubernetes operator or single-node binary.

### Is payload filtering as good as in a relational DB?

For straightforward equality and range filters, yes. For deeply nested JSON or complex joins, you still want a real DB upstream.

### How does quantization affect quality?

Qdrant supports scalar and product quantization to shrink vectors 4-32x with usually small recall loss. Worth enabling once your index passes a few hundred thousand vectors.

## Changelog

- **2026-05-26** — Refreshed install snippets and fact sheet; verified for 2026.
- **2025-02-15** — Initial directory listing.