AI in practice / FIELD NOTE

Step-by-Step: How to Use Kimi K2 to Get the $200 Claude Cowork Experience

Step-by-step: how to use Kimi K2 to get the $200 Claude Cowork experience

Translated from the 2026-01-17 version of the original

Introduction

Prerequisites

Before you start configuring, make sure that:

  1. Claude Code is installed and its basic configuration is done

  2. You have a Moonshot AI open platform account

  3. You have downloaded and installed Claude Cowork

Get a Kimi API Key

Step 1: Go to the Moonshot open platform

Go to the API platform and log in to your account.

Step 2: Create an API Key

On the API Keys management page, click to create a new API Key. I recommend choosing the default project.

key
[ key ]

Step 3: Save the API Key

Once it's created, store your API Key somewhere safe. For security reasons, the API Key is only shown once.

Install Claude Cowork

Download the app

Go to the Claude Cowork GitHub Releases page and download the version for your operating system:

  • macOS: download the .dmg file

  • Windows: download the .exe installer

  • Linux: download the .AppImage or .deb package

Install

Install the app following the standard procedure for your operating system.

Configure Claude Cowork to use Kimi K2

Claude Cowork configures which AI model it uses through environment variables. Because it is fully compatible with Claude Code's configuration, we can point it at the Kimi K2 model by setting environment variables.

This is the fast version with thinking capability, suitable for most scenarios.

macOS and Linux

Open a terminal and run the following commands:

bash
export ANTHROPIC_API_KEY="your-moonshot-api-key"
export ANTHROPIC_BASE_URL="https://api.moonshot.cn/v1"
export ANTHROPIC_MODEL="kimi-k2-thinking-turbo"

Then launch Claude Cowork from the same terminal window:

bash
open -a "Claude Cowork"

Windows

Open Command Prompt (CMD) and run:

set ANTHROPIC_API_KEY=your-moonshot-api-key
set ANTHROPIC_BASE_URL=https://api.moonshot.cn/v1
set ANTHROPIC_MODEL=kimi-k2-thinking-turbo

Then launch Claude Cowork from the same command-line window.

Or use PowerShell:

powershell
$env:ANTHROPIC_API_KEY="your-moonshot-api-key"
$env:ANTHROPIC_BASE_URL="https://api.moonshot.cn/v1"
$env:ANTHROPIC_MODEL="kimi-k2-thinking-turbo"

Option 2: use kimi-k2-turbo-preview

This is the fast version without thinking capability; it responds faster (up to 100 tokens/s).

macOS and Linux

bash
export ANTHROPIC_API_KEY="your-moonshot-api-key"
export ANTHROPIC_BASE_URL="https://api.moonshot.cn/v1"
export ANTHROPIC_MODEL="kimi-k2-turbo-preview"

Windows

set ANTHROPIC_API_KEY=your-moonshot-api-key
set ANTHROPIC_BASE_URL=https://api.moonshot.cn/v1
set ANTHROPIC_MODEL=kimi-k2-turbo-preview

Option 3: use kimi-k2-0905-preview

This is the standard version, strong at coding and suited to long-running scenarios.

macOS and Linux

bash
export ANTHROPIC_API_KEY="your-moonshot-api-key"
export ANTHROPIC_BASE_URL="https://api.moonshot.cn/v1"
export ANTHROPIC_MODEL="kimi-k2-0905-preview"

Windows

set ANTHROPIC_API_KEY=your-moonshot-api-key
set ANTHROPIC_BASE_URL=https://api.moonshot.cn/v1
set ANTHROPIC_MODEL=kimi-k2-0905-preview

Getting started

Once configuration is done, you can start using Claude Cowork! Here are a few usage tips:

Create a new session

  1. Create a new session in Claude Cowork

  2. Specify a working directory (this can be your project folder)

  3. Start talking to the AI

Common use cases

File management

请帮我整理这个文件夹,把所有图片移动到 images 子文件夹中

Learning and Q&A

请解释一下这段代码的工作原理

Session management

One of Claude Cowork's big advantages is its solid session management:

  • All conversation history is saved automatically in a local SQLite database

  • You can resume an earlier conversation at any time

  • Sessions you no longer need can be deleted

  • Each session can have its own working directory

Permission control

When the AI needs to perform a sensitive operation (such as deleting files or changing system settings), Claude Cowork pops up a confirmation dialog where you can choose:

  • Allow: perform the operation

  • Deny: skip the operation

  • View details: see exactly what the operation does

This gives you full control over what the AI does.

Summary

With this tutorial you've learned how to configure the Kimi K2 model for use in Claude Cowork. This combination gives you:

  • Strong AI capability: Kimi K2's code generation, agent and thinking abilities

  • A great user experience: Claude Cowork's graphical interface and session management

  • Flexible configuration: choose different model versions according to your needs

  • Cost control: avoid unexpected spending through budget settings

Remember these key points:

  1. Choose the right model for the task (thinking, turbo or standard)

  2. Set a daily spending budget to keep costs under control

  3. Keep monitoring usage and deal with anomalies promptly

  4. Make full use of Claude Cowork's session management and permission control features

Now go enjoy the productivity boost an AI assistant brings! If you run into problems, refer to the Moonshot AI official docs or the Claude Cowork GitHub repository.

SAME TOPIC

Prompts, demos and cases for the same model

USE CASE / CLAUDE01

Claude Fable 5.1 Use Cases

Explore 116 public cases showing Claude Fable 5.1 across games, frontend work, 3D, coding evaluations, media production, and workflows—with the evidence limits kept visible.

USE CASE / CLAUDE02

Claude Fable 5 Use Cases

See how creators put Claude Fable 5 to work across coding, long-running agents, visual design, and research.

USE CASE / KIMI03

Kimi K3 Use Cases

Explore real Kimi K3 cases across games, 3D, frontend motion, coding integrations, and evaluation limits.

NEXT STEP

Turn reading into your next step

Continue reading