Qdrant MCP Server

Qdrant MCP

A Model Context Protocol (MCP) server that exposes the full Qdrant vector database API as tools — collection management, advanced hybrid search, payload and vector editing, snapshots, and server observability. Not just store/find.

Table of Contents

• Overview
• Key Features
• Requirements
• Installation
  • PyPI
  • Docker
  • Claude Desktop bundle (.mcpb)
• Quick Start
• Connecting to Claude
  • Claude Desktop / Claude Code (local, stdio)
  • Remote (streamable-http)
  • Bring your own Qdrant (QDRANT_MCP_BYO)
• Configuration Reference
• Available Tools
• Read-Only Mode
• Architecture
• Security
• Development
• Versioning & Changelog
• License

Overview

Qdrant MCP is a thin MCP server that wraps the Qdrant API one-to-one: it registers a tool per Qdrant operation, validates the input with Pydantic, calls the official qdrant-client SDK, and returns the result. It never generates embeddings, never parses documents, and never decides how to chunk text — it is not a RAG system, on purpose. Whatever an LLM client wants to store or query, it brings its own vectors.

That focus is also what sets it apart from the official Qdrant MCP server, which exposes exactly two tools (store/find) and does embed documents for you. This server covers the rest of Qdrant's surface — everything under collections, points, search, payload, indexing, snapshots, and observability — so an LLM client can manage a Qdrant deployment end to end, not just push and pull memories through a narrow interface.

See ROADMAP.md for the full phase-by-phase design history, including every finding that shaped a decision (in Spanish).

Key Features

• Full collection & point lifecycle — create/update/delete collections, CRUD on points, scrolling, counting.
• Every Qdrant search mode — plain vector search, hybrid search (RRF/DBSF fusion with prefetch stages), grouped queries, recommend, discover, and pairwise distance matrices.
• Payload & vector editing — set/overwrite/delete/clear payload, payload indexes, facet counting, named (dense/sparse) vector management, atomic batch operations.
• Snapshots — per-collection and full-storage backup/restore.
• Observability — telemetry, Prometheus metrics endpoint, resource quotas, self-diagnosed issues.
• Opt-in tool surface — tools are grouped into toolsets you enable explicitly, so a client isn't handed 60+ overlapping tools by default.
• Read-only guard — one flag removes every mutating tool from the registry entirely.
• Bring-your-own-Qdrant mode — run a public endpoint with no database of your own; every caller supplies their own Qdrant, isolated by construction, protected by an SSRF guard.
• Resilient by default — every Qdrant call goes through retry-with-backoff and returns Qdrant's own error message on failure, never a generic exception.

Requirements

• Python 3.12+
• Qdrant server v1.19.0 or newer. Two tools (qdrant_collection_vector_create/qdrant_collection_vector_delete) depend on an endpoint that returns 404 on older Qdrant servers (verified against v1.13.6 and v1.15.1) — everything else works on older versions, but v1.19.0+ is the only version this project tests against.

Installation

PyPI

# Run without installing (recommended for Claude Desktop/Code configs)
uvx mcp-qdrant

# Or install into your environment
pip install mcp-qdrant

Docker

docker run --rm -p 8000:8000 \
  -e QDRANT_URL=http://host.docker.internal:6333 \
  -e QDRANT_MCP_TRANSPORT=streamable-http \
  -e QDRANT_MCP_HTTP_HOST=0.0.0.0 \
  -e QDRANT_MCP_SHARED_SECRET=<a long random secret> \
  ghcr.io/avaazquezz/qdrant-mcp:latest

The image only makes sense with streamable-http — stdio needs a client to own the process's stdin/stdout directly, which a detached container can't provide. The server binds 127.0.0.1 by default, so QDRANT_MCP_HTTP_HOST=0.0.0.0 is required for the port to be reachable from outside the container. Images are published on every tagged release as {version}, {major}.{minor}, and latest.

Claude Desktop bundle (.mcpb)

Download mcp-qdrant.mcpb from the latest release and double-click it. Claude Desktop installs the server via uv (resolving dependencies on your machine, no Python installation required) and prompts for Qdrant URL, API key, local path, toolsets, and read-only mode through its own settings form — no JSON to edit.

Quick Start

Point the server at a local Qdrant instance over stdio and confirm it's reachable:

QDRANT_URL=http://localhost:6333 mcp-qdrant

Once connected from an MCP client, a typical first exchange looks like:

1. qdrant_health_check — confirms the configured Qdrant instance is reachable before doing anything else.
2. qdrant_collection_create — {"collection_name": "docs", "vector_size": 4, "distance": "Cosine"}.
3. qdrant_points_upsert — {"collection_name": "docs", "points": [{"id": 1, "vector": [0.1, 0.2, 0.3, 0.4], "payload": {"title": "hello"}}]}.
4. qdrant_query — {"collection_name": "docs", "query_vector": [0.1, 0.2, 0.3, 0.4], "limit": 5}.

The full parameter shape of every tool is described in its own MCP schema — an LLM client reads those directly via tools/list; the table in Available Tools below is a human-readable summary of the same data.

Connecting to Claude

Claude Desktop / Claude Code (local, stdio)

claude_desktop_config.json (Claude Desktop) or .mcp.json (Claude Code):

{
  "mcpServers": {
    "qdrant": {
      "command": "uvx",
      "args": ["mcp-qdrant"],
      "env": {
        "QDRANT_URL": "http://localhost:6333",
        "QDRANT_MCP_TOOLSETS": "core,search"
      }
    }
  }
}

Or use the .mcpb bundle described in Installation — same result, no JSON to edit.

Remote (streamable-http)

For a server reachable over the network (e.g. added as a custom connector in Claude.ai) with a single, fixed backing Qdrant. QDRANT_MCP_SHARED_SECRET is required in this mode — the server refuses to start as streamable-http without one, to avoid serving an unauthenticated endpoint over the network (verified hands-on: an open streamable-http server is trivially usable by anyone with the URL).

QDRANT_URL=http://localhost:6333 \
QDRANT_MCP_TRANSPORT=streamable-http \
QDRANT_MCP_HTTP_HOST=0.0.0.0 \
QDRANT_MCP_SHARED_SECRET=<a long random secret> \
mcp-qdrant

In Claude.ai (Customize → Connectors → Add custom connector, verified hands-on against a real account): enter the server's HTTPS URL, then on the detected authentication screen choose "None" and add a Request header — Authorization → Bearer <the same secret>. (Authorization is used here because it's one of the two header names Claude.ai's custom-connector UI accepts without requiring Anthropic's manual approval of a custom name.)

Bring your own Qdrant (QDRANT_MCP_BYO)

A streamable-http deployment can run with no backing Qdrant of its own — every caller supplies their own Qdrant instance (their own Qdrant Cloud account, their company's self-hosted Qdrant, whatever) per request, instead of using one the operator hosts and pays for. Isolation between callers is automatic — each one talks to their own database — so there's no shared secret, no per-user account, and no data at rest on this server.

QDRANT_MCP_BYO=1 \
QDRANT_MCP_TRANSPORT=streamable-http \
QDRANT_MCP_HTTP_HOST=0.0.0.0 \
mcp-qdrant

Two request headers, reused for a different purpose than their name suggests — verified hands-on that Claude.ai's custom-connector "Request headers" UI rejects made-up header names outright unless Anthropic has approved them, so this reuses two pre-approved ones instead of inventing X-Qdrant-Url/X-Qdrant-Api-Key:

• Authorization (required) — your Qdrant URL, e.g. https://xyz.cloud.qdrant.io:6333. Sent verbatim, no Bearer prefix (unlike the personal-instance mode above, which uses the same header for a shared secret).
• x-api-key (optional) — your Qdrant API key, if your instance needs one.

In Claude.ai: Add custom connector → authentication "None" → add both as Request headers. Your Qdrant must be reachable from the public internet — see Security for what the SSRF guard rejects.

How it works, under the hood: every tool resolves its Qdrant client lazily, at call time, rather than once at startup. In BYO mode that client is a BYOQdrantClientProxy that reads the real AsyncQdrantClient from a contextvars.ContextVar, set per-request by the auth middleware — so none of the tool implementations need to know BYO mode exists. Clients are pooled in a bounded LRU cache (256 entries, keyed by URL + API key) so repeat callers don't pay a fresh TLS handshake on every call, with a 15-second grace period before an evicted client is closed so an in-flight request is never cut off mid-call.

Recommended setup: one MCP deployment, one Qdrant per project

This is the most common way to run this project: deploy the MCP once, in BYO mode, as a long-lived service — then, for each new project, spin up your own Qdrant and point a connector at it, without ever touching the MCP deployment again.

1. Deploy the MCP once, self-hosted (e.g. Docker behind a reverse proxy with TLS), in BYO mode as shown above. This never changes between projects.

2. Per project, run your own Qdrant with docker-compose, protected with its own API key (QDRANT__SERVICE__API_KEY — Qdrant's own auth, unrelated to this server):

   services:
     qdrant:
       image: qdrant/qdrant:latest
       restart: unless-stopped
       environment:
         QDRANT__SERVICE__API_KEY: ${QDRANT_API_KEY}
       volumes:
         - ./qdrant_storage:/qdrant/storage
   

   All data for that project lives in ./qdrant_storage, on your own server — the MCP deployment never stores or sees it beyond relaying each request.

3. Expose that Qdrant under its own public HTTPS domain (e.g. via Traefik/Let's Encrypt) — the SSRF guard rejects private/internal addresses, so it must be reachable from the public internet, not just from inside your server's Docker network.

4. Add one connector per project, pointing at the same MCP deployment but with different headers:

   • Claude.ai: a separate custom connector per project — Authorization = that project's Qdrant URL, x-api-key = its API key.
   • Claude Code (.mcp.json, remote HTTP server with custom headers):

     {
       "mcpServers": {
         "qdrant-project-x": {
           "type": "http",
           "url": "https://your-mcp.example.com/mcp",
           "headers": {
             "Authorization": "https://qdrant-project-x.example.com",
             "x-api-key": "${QDRANT_PROJECT_X_API_KEY}"
           }
         }
       }
     }
     

Adding a project is then just a new docker-compose up for its Qdrant plus a new connector — the MCP deployment itself is never redeployed or restarted.

Configuration Reference

All configuration is via environment variables, read once at startup — a misconfiguration fails immediately instead of surfacing later as a confusing tool error.

¹ QDRANT_URL and QDRANT_LOCAL_PATH are mutually exclusive; if neither is set, the client falls back to the qdrant-client SDK's own default of localhost:6333. ² In BYO mode the shared secret is optional — it adds an extra anti-bot gate on top of the per-caller isolation BYO already provides, rather than protecting shared data. BYO mode also requires QDRANT_MCP_TRANSPORT=streamable-http and is mutually exclusive with QDRANT_URL/QDRANT_LOCAL_PATH (there is nothing "backing" to point at).

Available Tools

Tools are grouped into toolsets, enabled via QDRANT_MCP_TOOLSETS (comma-separated). Only core is enabled by default — the rest are explicit opt-ins, so a client isn't handed every tool at once:

• core — collection and point CRUD, plus qdrant_query and the qdrant_health_check smoke test. Enough for a fully working MCP on its own.
• search — everything beyond plain vector search: batched/grouped queries, recommend, discover, and pairwise distance matrices.
• payload — payload editing and indexing, named-vector management, atomic batch point operations.
• snapshots — collection and full-storage backup/restore.
• observability — telemetry, metrics, quotas, and self-diagnosed issues.
• admin is a reserved toolset name with no registered tools — cluster/shard administration was scoped out (see ROADMAP.md, Fase 5): its most useful capability, real resharding, only exists on Qdrant Cloud, and the rest only matters for a distributed deployment. Setting QDRANT_MCP_TOOLSETS=admin is valid but registers nothing.

The table below is generated directly from the live tool registry — run uv run python scripts/gen_tools_doc.py after adding or changing a tool to keep it in sync (CI fails the build if it drifts).

Read-Only Mode

Setting QDRANT_MCP_READ_ONLY=1 removes every tool whose destructiveHint isn't explicitly false (i.e. anything that isn't readOnlyHint=true) from the registry itself — a client calling tools/list never sees them, rather than seeing them and having calls rejected. This is the mechanism to hand an LLM client safe, read-only access to a Qdrant deployment: point it at your database, set the flag, and there is no code path left for it to write anything.

Architecture

• Single shared client. One AsyncQdrantClient (or, in BYO mode, a proxy — see below) is built once at startup and closed over by every tool. Tools resolve it lazily at call time rather than caching anything from it at registration.
• One registration choke point. Every tool module calls into a single ToolRegistry, which is where the toolset filter and the read-only guard are both enforced — a tool can't bypass either by registering itself differently.
• Resilience. Every Qdrant call is wrapped with tenacity-based retry and backoff (the qdrant-client SDK itself only exposes a timeout, not retries), and translated into an MCP ToolError carrying Qdrant's own error message instead of a generic failure — so "collection not found" reads as exactly that.
• stdio-safe logging. All logging goes to stderr, configured before the MCP server object is even constructed — writing to stdout under the stdio transport would corrupt the JSON-RPC message framing.
• BYO mode's indirection. In QDRANT_MCP_BYO mode, the "client" every tool holds is a proxy that reads the real, per-caller AsyncQdrantClient out of a context variable set by request middleware — see Bring your own Qdrant for the full mechanism.

Security

• Mandatory authentication for the personal streamable-http mode. The server refuses to start without QDRANT_MCP_SHARED_SECRET — verified hands-on that an unauthenticated instance is trivially usable by anyone with the URL.
• SSRF protection in BYO mode. Since a BYO deployment connects to whatever URL an untrusted public caller supplies — from a host that may also run other services on internal networks — every URL is checked before use: only http/https, DNS-resolved, and rejected if any resolved address is private, loopback, link-local, or the cloud metadata address. This check re-resolves on every request rather than caching a prior result, specifically to defeat DNS rebinding.
• Input validation everywhere. Every tool's input is a Pydantic model — malformed input is rejected with a structured error before it reaches the Qdrant SDK.
• No secrets at rest, no secrets logged. BYO mode holds no long-lived credentials; API keys passed via x-api-key live only as long as their pooled client connection.

Development

git clone https://github.com/avaazquezz/Qdrant-MCP.git
cd Qdrant-MCP
uv sync --dev
uv run pre-commit install

uv run ruff check .            # lint
uv run ruff format --check .   # formatting
uv run mypy .                  # type checking (strict)
uv run pytest                  # unit tests
uv run pytest -m integration   # integration tests — needs a running Qdrant (CI runs one as a Docker service)

If you add or change a tool, regenerate the table in Available Tools rather than hand-editing it:

uv run python scripts/gen_tools_doc.py          # regenerate
uv run python scripts/gen_tools_doc.py --check  # verify, no changes (what CI runs)

Versioning & Changelog

This project follows Semantic Versioning. See CHANGELOG.md for release notes and ROADMAP.md for the phased design history behind each version.

License

MIT