Qdrant MCP for Zed

A Zed extension to access Qdrant vector database directly in Zed using the Model Context Protocol (MCP). Works with any OpenAI-compatible embedding server (llama.cpp, Ollama, vLLM, LM Studio, and more).

Table of Contents

• Prerequisites
• Quick Start
• Configuration
• Compatible Embedding Servers
• Recommended Embedding Models
• Available Tools
• Installing Extensions
• Installation Location
• Development
• Resources

Prerequisites

Qdrant

You need a running Qdrant instance. Options:

Qdrant Cloud (recommended for production)

1. Sign up at https://cloud.qdrant.io
2. Create a cluster
3. Copy URL and API key to settings (default port is added automatically: :6334 for gRPC, :6333 for REST)

Local Qdrant with Docker

# Quick start (data lost when container stops)
docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

# With persistent storage (recommended)
docker run -d --name qdrant \
  -p 6333:6333 -p 6334:6334 \
  -v qdrant_storage:/qdrant/storage \
  --restart unless-stopped \
  qdrant/qdrant

Manage the container:

docker stop qdrant    # stop
docker start qdrant   # start again (data persists)
docker logs qdrant    # view logs

Local Qdrant binary on macOS (no Docker)

Download the latest binary from Qdrant releases:

• Apple Silicon (M1/M2/M3/M4): qdrant-aarch64-apple-darwin.tar.gz
• Intel: qdrant-x86_64-apple-darwin.tar.gz

# Extract and run
tar -xzf qdrant-*.tar.gz
./qdrant

Qdrant will start on the default ports (6333 HTTP, 6334 gRPC).

Embedding Server

You need any server that exposes an OpenAI-compatible /v1/embeddings endpoint. See Compatible Embedding Servers for options.

Quick Start

1. Start Qdrant (cloud or local)
2. Start an embedding server (see Compatible Embedding Servers)
3. Install this extension in Zed (see Installing Extensions)
4. Add settings to ~/.config/zed/settings.json:

   {
     "context_servers": {
       "mcp-server-qdrant": {
         "settings": {
           "embedding_server_url": "http://localhost:8080",
           "embedding_model": "snowflake-arctic-embed2"
         }
       }
     }
   }

5. Use qdrant-store to save text and qdrant-find to search semantically

Configuration

Configure in ~/.config/zed/settings.json:

{
  "context_servers": {
    "mcp-server-qdrant": {
      "settings": {
        "qdrant_url": "https://xxx.cloud.qdrant.io",
        "qdrant_api_key": "your-api-key",
        "collection_name": "zed_collection",
        "embedding_server_url": "http://localhost:8080",
        "embedding_api_key": "",
        "embedding_model": "snowflake-arctic-embed2",
        "transport_protocol": "grpc",
        "enabled_tool_groups": "core"
      }
    }
  }
}

Setting	Default	Description
qdrant_url	—	Qdrant server URL
qdrant_api_key	—	API key for Qdrant Cloud
collection_name	—	Vector collection name (created automatically)
embedding_server_url	—	URL of any OpenAI-compatible embedding server
embedding_api_key	—	Optional API key for the embedding server
embedding_model	—	Model name passed to the embedding server
transport_protocol	grpc	Transport protocol: "grpc" (port 6334) or "rest" (port 6333)
enabled_tool_groups	core	Comma-separated list of tool groups to enable (see Tool Groups)

Transport Protocol

The extension supports both gRPC and REST for communicating with Qdrant:

• gRPC (default) — Binary protocol, better performance for large data. Uses port 6334.
• REST — JSON-based, easier to debug. Uses port 6333.

When running Qdrant locally with Docker, expose both ports to allow switching:

docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

Set "transport_protocol": "rest" in your settings to use REST. The correct default port is applied automatically when qdrant_url doesn't specify one.

Compatible Embedding Servers

Any server that exposes an OpenAI-compatible /v1/embeddings endpoint will work. Configure embedding_server_url and embedding_model accordingly.

Server	Default URL	Example embedding_model	Start Command
llama.cpp	http://localhost:8080	snowflake-arctic-embed2	llama-server -m model.gguf --embeddings --port 8080
Ollama	http://localhost:11434	nomic-embed-text	ollama serve (then ollama pull nomic-embed-text)
vLLM	http://localhost:8000	BAAI/bge-m3	vllm serve BAAI/bge-m3
LM Studio	http://localhost:1234	nomic-embed-text	Start server from LM Studio UI
LocalAI	http://localhost:8080	bert-cpp-minilm-v6	local-ai run bert-cpp-minilm-v6

Recommended Embedding Models

Model	Dimensions	Max Tokens	Best For
snowflake-arctic-embed2	768	8192	General-purpose, multilingual retrieval
nomic-embed-text v2	768	8192	Multilingual, MoE architecture
nomic-embed-text v1.5	768	8192	Lightweight, GGUF-ready
BAAI/bge-m3	1024	8192	Multilingual, dense + sparse retrieval
mxbai-embed-large	1024	512	High accuracy, English-focused
jina-embeddings-v4	2048	8192	Multimodal (text + images)

Tool Groups

Tools are organized into groups to reduce context window usage. By default, only the core group is enabled (~450 tokens). Enabling all 27 tools consumes ~5,500 tokens of context.

Group	Tools	Description
core	3	Semantic store, find, and list collections (default)
collections	5	Create, delete, info, exists, update collections
points	5	Upsert, get, delete, count, scroll points
search	6	Vector search, recommend, query, batch, groups, discover
payload	4	Set/delete payload, create/delete indexes
admin	4	Aliases and snapshots

Enable groups in your settings with a comma-separated list:

{
  "context_servers": {
    "mcp-server-qdrant": {
      "settings": {
        "enabled_tool_groups": "core,collections,search"
      }
    }
  }
}

Use "all" to enable every group:

"enabled_tool_groups": "all"

Available Tools

Semantic Operations

• qdrant-store - Store text with automatically generated embeddings
• qdrant-find - Search for semantically similar texts

Collection Management (6 tools)

• qdrant-list-collections - List all collections
• qdrant-create-collection - Create a new vector collection
• qdrant-delete-collection - Delete a collection
• qdrant-get-collection-info - Get collection details
• qdrant-update-collection - Update collection parameters (HNSW, optimizer, etc.)
• qdrant-collection-exists - Check if a collection exists

Point Operations (5 tools)

• qdrant-upsert-points - Insert or update points with vectors and payloads
• qdrant-get-points - Retrieve points by ID
• qdrant-delete-points - Delete points by ID or filter
• qdrant-count-points - Count points in a collection
• qdrant-scroll-points - Paginate through points

Search & Query (6 tools)

• qdrant-search - Raw vector similarity search
• qdrant-recommend - Find similar points by example IDs (no vector needed)
• qdrant-query - Universal query endpoint (fusion, re-ranking, nested prefetch)
• qdrant-search-batch - Execute multiple searches in one request
• qdrant-search-groups - Search with results grouped by a payload field
• qdrant-discover - Context-based exploration with positive/negative pairs

Payload & Index Management (4 tools)

• qdrant-set-payload - Set or update payload fields on existing points
• qdrant-delete-payload - Delete specific payload keys from points
• qdrant-create-index - Create a payload field index for faster filtered searches
• qdrant-delete-index - Delete a payload field index

Aliases (2 tools)

• qdrant-list-aliases - List all aliases or aliases for a specific collection
• qdrant-update-aliases - Create, rename, or delete collection aliases

Snapshots (2 tools)

• qdrant-list-snapshots - List snapshots for a collection
• qdrant-create-snapshot - Create a snapshot for backup

Installing Extensions

You can search for extensions by launching the Zed Extension Gallery by pressing cmd-shift-x, opening the command palette and selecting zed: extensions, or by selecting "Zed > Extensions" from the menu bar.

Here you can view the extensions that you currently have installed or search and install new ones.

Installation Location

• On macOS, extensions are installed in ~/Library/Application Support/Zed/extensions.
• On Linux, they are installed in either $XDG_DATA_HOME/zed/extensions or ~/.local/share/zed/extensions.
• On Windows, the directory is %LOCALAPPDATA%\Zed\extensions.

This directory contains two subdirectories:

• installed, which contains the source code for each extension.
• work, which contains files created by the extension itself, such as downloaded language servers.

Development

See docs/DEVELOPMENT.md for build instructions, test setup, and dependency management.

Resources

Repository

• Qdrant MCP Server Extension for Zed Repository on GitHub

Zed

• Zed Docs – Installing Extensions
• Zed Docs – Developing Extensions
• Zed Docs – MCP Server Extensions
• Zed Docs – Model Context Protocol (MCP)
• Zed Industries/Extensions GitHub Repository

Qdrant

• Qdrant API Reference
• Qdrant API & SDK documentation
• Qdrant Rust client on GitHub
• Qdrant Open API definition on GitHub
• Qdrant protobuf definition on GitHub
• Official Qdrant Model Context Protocol (MCP) server implementation
• qdrant-tech profile on npmjs
• Qdrant JavaScript SDK (REST + gRPC)

Embedding Servers

• llama.cpp on GitHub
• llama.cpp server documentation
• Ollama
• vLLM
• LM Studio
• LocalAI