[説明] このページは機械翻訳された日本語版です。内容に誤りがございましたら、報告していただけると助かります。

CLI と Agent 連携クイックスタート

このガイドでは、ローカル環境で Zilliz CLI と Agent 連携をセットアップする方法を説明します。セットアップ後は、自然言語で Agent から Zilliz Cloud を操作したり、CLI をターミナル・スクリプト・CI ワークフローで直接利用したりできます。

Installation

Before you start, ensure that you have:

• A Zilliz Cloud account.

• Claude Code, if you want to use the Claude Code Plugin.

• Node.js, if you want to install Zilliz Skill.

Install Claude Code Plugin

Use the Claude Code Plugin if you want to operate Zilliz Cloud directly from Claude Code.

1

Run Claude Code

> claude

2

Open the plugin marketplace

/plugin

3

Find and install the Zilliz Plugin

Go to the Discover tab and search for zilliz. Select the zilliz plugin to install it.

4

Run the quickstart wizard. The wizard guides you through CLI installation, authentication, cluster connection, and first operations.

/zilliz:quickstart

Install Zilliz Skill for common agent frameworks

If your coding agent, such as Codex, Gemini CLI, Cursor, or another skill-compatible agent, supports agent skills, install Zilliz Skill as follows:

npx skills add zilliztech/zilliz-skill

This command will prompt you to choose the target agent framework and installation scope.

Install Zilliz CLI

The Zilliz CLI is the base command-line tool used by the Plugin and Skill.

1

Install Zilliz CLI.

macOS / Linux

curl -fsSL https://zilliz.com/cli/install.sh | bash

Windows

irm https://zilliz.com/cli/install.ps1 | iex

Verify installation:

zilliz --version

2

Authenticate.

Authenticate with your Zilliz Cloud account:

zilliz login

This opens a browser for authentication. After login, your credentials are stored locally.

When to use CLI, Plugin, or Skill

Use these tools when you need to:

• Develop and test manually from your local environment.

• Write automated operations scripts for repeatable workflows.

• Enable your agent to call Vector Database or Vector Lakebase services automatically.

Tool comparison

The Claude Code Plugin, Zilliz Skill, and Zilliz CLI should cover the same main capabilities. Choose based on workflow, not feature scope.

	Claude Code Plugin	Zilliz Skill	Zilliz CLI
Best for	Claude Code natural-language workflows	Skill-compatible coding agents	Terminal use, scripts, and CI
Setup	/zilliz:quickstart	npx skills add zilliztech/zilliz-skill	Install script + zilliz login
Natural language	Yes	Yes	No
Automation	Agent-assisted	Agent-assisted	Script-first
Structured output	Agent-readable responses	Agent-readable responses	--output json for scripts

Supported capabilities

The following table explains the capabilities of the CLI, Plugin and Skill.

Area	What You Can Do
Clusters	Create, delete, suspend, resume, modify
Collections	Create with custom schema, load, release, rename, drop
Vectors	Search, query, insert, upsert, delete, hybrid search
Indexes	Create (AUTOINDEX), list, describe, drop
Databases	Create, list, describe, drop
Users & Roles	RBAC setup, privilege management
Backups	Create, restore, export, policy management
Import	Bulk data import from S3/GCS/Azure Blob Storage
Partitions	Create, load, release, manage
Monitoring	Cluster status, collection stats, load states
Projects	Project and region management
Billing	Usage queries, invoices

What you can ask your agent to do

After installation, describe the task directly. Your agent should translate the request into the corresponding Zilliz CLI commands. The following examples show how a natural-language request maps to the CLI commands your agent is expected to run.

• List my clusters and show which one is currently active.

  Expected CLI commands:

  zilliz cluster list
  zilliz context current

• Create a collection for product embeddings with a 768-dimensional vector field.

  Expected CLI command:

  zilliz collection create --name product_embeddings --dimension 768

• Import data from S3 into my collection and check the import job status.

  Expected CLI command:

  zilliz import start --cluster-id <cluster-id> --collection product_embeddings --body '{"files": [["s3://bucket/path/data.json"]]}'

• Create a backup for my production cluster.

  Expected CLI command:

  zilliz backup create --cluster-id <cluster-id>

• Search my collection with a metadata filter and return the top 10 results.

  Expected CLI command:

  zilliz vector search --collection product_embeddings --data '[[0.1, 0.2, 0.3]]' --filter 'age > 20' --limit 10 --output-fields '["name", "age"]'

• Create a role with read-only access to the analytics collection.

  Expected CLI commands:

  zilliz role create --role analytics_readonly
  zilliz role grant-privilege --role analytics_readonly --object-type Collection --object-name analytics --privilege Search
  zilliz role grant-privilege --role analytics_readonly --object-type Collection --object-name analytics --privilege Query