Connect Claude Code & Codex to DeepSeek V4 Pro: Config, Cost Analysis & Verification

Overview

This article explains how to connect the popular AI coding agents Claude Code and Codex to domestic Chinese models, using DeepSeek V4 Pro as the primary example. Domestic users often face barriers with Anthropic and OpenAI services including regional availability, network connectivity, payment methods, and rising token costs for long-running agent tasks. Integrating domestic models preserves the clients' file access, terminal, tool-calling, and task-loop capabilities while routing model requests to a local provider.

Important boundary: client capabilities and model capabilities are not fully equivalent. Even with API compatibility, third-party models may differ in tool calling, reasoning messages, context compression, vision input, or built-in tools. Code, documents, and conversations sent to the new provider fall under their data processing terms; verify organizational policies before handling sensitive code.

Prerequisites

Install Clients

Both Claude Code and Codex offer desktop apps and CLIs. Desktop apps are more intuitive; CLIs suit project-directory work and automation.

Codex desktop: https://chatgpt.com/codex/ Claude Code desktop: https://claude.com/download Codex CLI install (macOS/Linux):

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Or via npm: npm install -g @openai/codex. Verify with codex --version. Start in a project directory: cd /path/to/project && codex. First run may prompt login; exit because we'll switch to DeepSeek API.

Claude Code CLI install (macOS/Linux/WSL): curl -fsSL https://claude.ai/install.sh | bash Windows PowerShell: irm https://claude.ai/install.ps1 | iex. Or with Node.js 22+: npm install -g @anthropic-ai/claude-code (do not use sudo). Verify with claude --version and claude doctor. Start: cd /path/to/workspace && claude. Confirm codex --version and claude --version output before proceeding.

Create DeepSeek API Key

Register and log in to DeepSeek Open Platform.

Add a small balance.

Go to API Keys page.

Create a new API Key.

Copy immediately and store securely (password manager, keychain, secret manager). Do not write to plain notes, commit to Git, or expose in screenshots.

Placeholder used in examples: <your DeepSeek API Key> (replace without angle brackets).

Manage Multiple Configurations with CC Switch

CC Switch is an open-source third-party tool that manages provider configurations for Claude Code, Claude Desktop, and Codex. It reads/writes local configs and stores credentials. Download from https://ccswitch.io. If you only need one provider or prefer not to use a third-party credential manager, skip to manual configuration sections.

In CC Switch's Codex panel, click "+", select DeepSeek, enter API Key, save, and enable. Restart Codex fully and test connection. For Claude Code CLI, similar steps in the Claude Code panel. For Claude Desktop, model mapping is required because DeepSeek model names don't match Claude's native Sonnet/Opus/Haiku roles. Enable "Model Mapping" and map roles to actual DeepSeek models. Then enable local routing (listens on http://127.0.0.1:15721), restart Claude Desktop. Notes: mapping mode requires CC Switch and local routing running; direct mode doesn't. Claude Desktop doesn't support hot reload; full restart needed. Linux not supported for Claude Desktop config in CC Switch.

Codex CLI & Desktop App: Connect to DeepSeek

Method 1: CC Switch (covered above)

Method 2: DeepSeek Official Script

macOS/Linux:

bash <(curl -fsSL https://cdn.deepseek.com/api-docs/codex-deepseek-setup-en.sh)

Windows PowerShell:

irm https://cdn.deepseek.com/api-docs/codex-deepseek-setup-en.ps1 | iex

Script steps: select deepseek-v4-pro, enter API Key, wait for config and syntax validation, fully restart Codex. Script backs up original config.toml to ~/.codex/backup-deepseek/, generates ~/.codex/models.json, adds DeepSeek provider to config.toml, preserves MCP servers and project trust, validates TOML/JSON syntax, provides restore option. Security note: curl | bash executes remote script; in high-security environments, download, inspect, then run manually.

Method 3: Manual Configuration

Step 1: Save official model catalog to ~/.codex/models.json (copy from DeepSeek Codex integration page to avoid missing metadata like context windows, reasoning tiers, tool formats).

Step 2: Set API Key environment variable.

macOS/Linux: export DEEPSEEK_API_KEY="<your DeepSeek API Key>" Windows PowerShell: $env:DEEPSEEK_API_KEY="<your DeepSeek API Key>" Step 3: Edit ~/.codex/config.toml:

model = "deepseek-v4-pro"