Complete Guide to Claude Code Desktop: Direct Access in China + Chinese Localization + DeepSeek Integration

Introduction

Claude Code recently launched its Desktop version, offering a much friendlier graphical interface compared to the previous command-line (CLI) version. However, for users in China, using Claude's official models directly not only requires a VPN but also comes with hefty costs. This article provides a detailed walkthrough on how to install Claude Code Desktop and connect it to domestic models like DeepSeek through third-party tools — covering account-free usage, Chinese localization, and custom Skill configuration.

Downloading and Installing Claude Code Desktop

Claude Code's Product Positioning and Evolution

Claude Code is an AI programming assistant developed by Anthropic, initially released as a command-line interface (CLI) tool positioned as an "AI coding companion in your terminal." Unlike traditional IDE plugins (such as GitHub Copilot or Cursor), Claude Code takes an Agent-based approach — it doesn't just complete code but can understand project structures, execute terminal commands, and read/write files, delivering an experience closer to "autonomous programming." The Desktop release marks the product's transition from a niche developer tool to one accessible to a broader developer audience, significantly lowering the barrier to entry.

Downloading the Installer from the Official Website

First, visit the official Claude Code website. On the homepage, find the "Get Claude Code" button. Hover over it to see multiple usage options — Desktop is listed first. Click it to go to the download page, where the system will automatically detect your operating system and provide the matching version (Windows/Mac).

Once downloaded, double-click the executable to install. The process is straightforward — just click "Next" all the way through. After installation, you'll see a login screen offering sign-in options like Google accounts.

Account-Free Usage (Enabling Developer Mode)

If you don't want to register or log in, you can bypass this by enabling Developer Mode:

1. Click the hamburger menu (three horizontal lines) in the top-left corner
2. Select "Help"
3. Find the third option, scroll down to "Enable Developer Mode"
4. Click to confirm, then wait for Claude Code to restart

After restarting, the interface will look slightly different. Don't do anything else just yet — you'll need to install a companion tool first.

Security Notes on Developer Mode

With Developer Mode enabled, Claude Code skips Anthropic's official authentication process and instead relies on user-configured third-party APIs. This means your conversation data no longer passes through Anthropic's servers but is sent directly to third-party model providers (such as DeepSeek). From a privacy perspective, this can actually be an advantage — data stays on domestic servers, complying with data localization requirements. However, note that in Developer Mode, you won't benefit from Anthropic's official safety filtering and content moderation mechanisms, so you'll need to take responsibility for the compliance of model outputs yourself.

Connecting to DeepSeek via CC Switch

What Is CC Switch?

CC Switch is a software tool for unified management of AI tool workflows. Its core function is helping users switch between different AI models. It essentially acts as a local routing proxy, allowing Claude Code Desktop to call third-party model APIs like DeepSeek, enabling direct access from China without a VPN.

Technical Principles of the Local Routing Proxy

The "local routing" implemented by CC Switch is essentially an API Reverse Proxy. When Claude Code makes an API request, it would normally be sent to Anthropic's official servers. CC Switch starts a proxy service locally (typically listening on a localhost port), intercepts these requests, and forwards them to the user-specified third-party model API endpoint (e.g., DeepSeek's api.deepseek.com). The advantage of this approach is that it doesn't require modifying Claude Code's source code — simply changing the target address of API requests enables seamless model switching. This is also the common principle behind the "one-click source switching" solutions popular in the open-source community.

Important: CC Switch must be version 3.15.0 or above, as older versions don't support Claude Code Desktop configuration. If you have an older version installed, make sure to update it.

Installation is equally straightforward — double-click the installer and follow the prompts.

Configuring the DeepSeek API Key

After installation, follow these steps to configure:

1. Enable local routing: Click the "Settings" button on the left → Click "Routing" → Find "Local Routing" → Click the dropdown arrow to enable it
2. Select Desktop mode: Return to the main interface and select "Desktop" at the top (you'll see a small computer icon)
3. Add a model: Click the "Add" button on the right → Select DeepSeek
4. Enter your API Key: Go to the API documentation page on DeepSeek's official website, create a new API Key in the Key management section, copy it, and paste it into CC Switch
5. Enable context extension: It's recommended to check the 1M context size option to give Claude Code a larger context window
6. Start the service: After clicking "Add," make sure to click the "Start" button — otherwise the configuration won't take effect

DeepSeek's Technical Features and Cost Advantages

DeepSeek is a large language model series developed by DeepSeek AI. Among its models, DeepSeek-V3 and DeepSeek-R1 have demonstrated capabilities approaching GPT-4 and Claude 3.5 across multiple benchmarks. DeepSeek's core technical advantages include: using MoE (Mixture of Experts) architecture to significantly reduce inference costs; supporting ultra-long context windows (up to 128K tokens); and particularly strong performance in code generation and mathematical reasoning. Its API pricing is roughly 1/10 to 1/20 of Claude's official models, bringing the monthly cost of heavy AI coding assistant usage down from hundreds of yuan to just tens of yuan or even less.

Once started, you can minimize CC Switch (clicking the close button actually minimizes it to the system tray).

Completing the Integration in Claude Code Desktop

Return to Claude Code Desktop:

1. Click the hamburger menu in the top-left corner
2. You'll now see a new "Developer" option
3. Click "Developer" → "Configure Third-Party"
4. The information should already be filled in (CC Switch automatically syncs the configuration)
5. Click "Apply" and restart

After restarting, you'll enter the Claude Code main interface, where the model list will show Flash and Pro versions along with the 1M context version. Select the desired model and you're ready to start chatting.

Chinese Localization for Claude Code Desktop

By default, the Claude Code Desktop interface is entirely in English, including menus and buttons. The community has created Chinese localization patches supporting both Windows and Mac.

Localization Steps

1. Download the localization patch archive (includes both Windows and Mac versions)
2. Extract it and find the .bat file (Windows), then double-click to run it
3. You'll see five options:
   • 1: Install Simplified Chinese
   • 2: Install Traditional Chinese
   • 3: Install Traditional Chinese (Hong Kong)
   • 4: Restore original / Uninstall patch
   • 5: Exit
4. Type 1 and press Enter to install Simplified Chinese
5. The script will automatically close Claude Code and apply the localization
6. After completion, Claude Code will restart automatically

After restarting, you'll see the interface is fully in Chinese, including menu items like "New Task," "Project Plan," "Tasks," "Customize," and more.

Custom Skill Upload and Usage

Design Philosophy of the Skill System

Claude Code's Skill system is essentially a structured prompt template framework. Each Skill is typically defined by a Markdown file containing role definitions, execution steps, output format instructions, and more. This design draws from the "Chain" and "Agent" concepts in frameworks like LangChain, breaking complex tasks into reusable skill modules. Users can combine different Skills to build their own workflows — for example, "Requirements Analysis Skill + Code Generation Skill + Unit Testing Skill" forms a complete development pipeline. The community already offers a large number of open-source Skills ready for direct use.

Differences Between CLI and Desktop Versions

In the CLI version, Skill files are simply placed in the .claude/skills/ directory. The Desktop version, however, requires uploading and installing Skills as compressed archives, so the process is slightly different.

How to Upload Skills

1. Create a new task
2. Click the "+" button → "Skills" → "Manage Skills"
3. Click "Add Skill" → "Upload Skill"
4. Supported file formats: .md files, archives (.zip), .skill files
5. Compress your Skill folder into a zip file and drag-and-drop to upload

How to Invoke Skills

After uploading, type a forward slash / in the chat input, select "Skills," then choose the corresponding Skill name to invoke it. For example, select a creative generation Skill, enter your requirements, and Claude Code will generate content following the rules defined in that Skill.

Summary

Through the configuration steps above, we've achieved:

• Installed Claude Code Desktop: Downloaded from the official site with a straightforward installation
• Account-free usage: Bypassed login by enabling Developer Mode
• Connected to domestic models: Configured DeepSeek API via CC Switch for direct access from China
• Chinese localization: One-click interface localization using a community patch
• Custom Skills: Uploaded and used via compressed archives

The core advantage of this setup is the dramatic cost reduction — DeepSeek's API pricing is far lower than Claude's official models, while CC Switch's local routing enables direct access from China without a VPN. For developers who want to experience the Claude Code workflow on a limited budget, this is an extremely practical alternative.

From an industry trend perspective, this "frontend tool + swappable backend model" architecture is becoming the mainstream pattern for AI development tools. Developers are no longer locked into a specific model provider and can flexibly switch based on task type, budget, and network environment. As domestic large language models continue to improve and API prices drop further, the practical value of these localized solutions will only continue to grow.