Run Claude Code CLI with the DeepSeek API: A Lower-Cost Terminal Coding Assistant Setup
TL;DR
Claude Code CLI can run against DeepSeek's Anthropic-compatible API. The essential setup is to point ANTHROPIC_BASE_URL at https://api.deepseek.com/anthropic, put your DeepSeek API key in ANTHROPIC_AUTH_TOKEN, and map Claude Code's main and lightweight model variables to DeepSeek models.
Last reviewed: 2026-06-02. Maintained by the ToolOrbit Editorial Team.
Claude Code is Anthropic's terminal-based coding assistant. It can read code inside a project directory, edit files, run commands, debug issues, refactor modules, generate tests, and explain complex code. For frequent coding work, it feels less like a chatbot and more like an AI teammate that can participate directly in engineering tasks.
The practical concern is cost. If you use Claude Code heavily, API usage can add up quickly. That leads to a common question:
Can you keep the Claude Code CLI workflow while switching the backend model provider to DeepSeek?
Yes.
DeepSeek provides an Anthropic-compatible API and documents a Claude Code configuration path. That means Claude Code CLI can remain your terminal interface while requests are routed to DeepSeek's model service.
1. What you need first
Before starting, make sure you have:
- Node.js 18 or newer
- Claude Code CLI
- A DeepSeek API key
- A project directory where you can run terminal commands
If Claude Code is not installed yet, run:
npm install -g @anthropic-ai/claude-code
Then verify the installation:
claude --version
Create your DeepSeek API key in the DeepSeek Platform dashboard.
API keys are sensitive. Do not commit them to Git and do not share screenshots that expose them.
2. The core configuration
Claude Code supports environment variables for the API endpoint, auth token, and default models.
The key variables are:
ANTHROPIC_BASE_URL
ANTHROPIC_AUTH_TOKEN
ANTHROPIC_MODEL
ANTHROPIC_DEFAULT_OPUS_MODEL
ANTHROPIC_DEFAULT_SONNET_MODEL
ANTHROPIC_DEFAULT_HAIKU_MODEL
CLAUDE_CODE_SUBAGENT_MODEL
CLAUDE_CODE_EFFORT_LEVEL
The most important one is:
ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
This tells Claude Code to send Anthropic-format requests to DeepSeek's compatible endpoint.
DeepSeek's Claude Code setup points to this model split:
Main model: deepseek-v4-pro[1m]
Fast model: deepseek-v4-flash
API endpoint: https://api.deepseek.com/anthropic
deepseek-v4-pro[1m] is better suited to complex coding tasks and long-context projects. deepseek-v4-flash is a better fit for subagents, fast responses, and lighter work.
3. macOS / Linux / WSL setup
On macOS, Linux, or WSL, run:
export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_AUTH_TOKEN=<your DeepSeek API key>
export ANTHROPIC_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-v4-pro[1m]
export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash
export CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash
export CLAUDE_CODE_EFFORT_LEVEL=max
Then enter your project directory:
cd /path/to/your-project
claude
If you want the variables to persist, add them to ~/.zshrc or ~/.bashrc.
For example:
nano ~/.zshrc
Reload the shell config:
source ~/.zshrc
From a security perspective, avoid putting API keys in files that are synced, shared, or committed.
4. Windows PowerShell setup
On Windows, run this in PowerShell:
$env:ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
$env:ANTHROPIC_AUTH_TOKEN="<your DeepSeek API key>"
$env:ANTHROPIC_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek-v4-pro[1m]"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-v4-flash"
$env:CLAUDE_CODE_SUBAGENT_MODEL="deepseek-v4-flash"
$env:CLAUDE_CODE_EFFORT_LEVEL="max"
Then enter your project directory:
cd C:\your-project
claude
PowerShell $env: assignments only apply to the current window. After closing that window, you need to set them again. You can use user-level environment variables for persistence, but still handle API keys carefully.
5. Recommended approach: local project settings
If only one project should use DeepSeek, use Claude Code's local settings file instead of changing your global shell environment.
Create this file in the project:
.claude/settings.local.json
Add:
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
"ANTHROPIC_AUTH_TOKEN": "your DeepSeek API key",
"ANTHROPIC_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash",
"CLAUDE_CODE_SUBAGENT_MODEL": "deepseek-v4-flash",
"CLAUDE_CODE_EFFORT_LEVEL": "max"
}
}
This is useful for project-level routing. Project A can use DeepSeek while Project B continues to use native Claude. The two setups do not have to interfere with each other.
If .claude/settings.local.json contains an API key, do not commit it. Check that .gitignore excludes local Claude settings.
6. Why ANTHROPIC_AUTH_TOKEN instead of ANTHROPIC_API_KEY
This is an easy place to make a mistake.
Claude Code supports multiple authentication-related environment variables. DeepSeek's Claude Code integration recommends:
ANTHROPIC_AUTH_TOKEN
Claude Code documentation says ANTHROPIC_AUTH_TOKEN is sent as a Bearer token. ANTHROPIC_API_KEY is commonly sent as X-Api-Key.
For Claude Code + DeepSeek, prefer the DeepSeek-documented option:
ANTHROPIC_AUTH_TOKEN=<your DeepSeek API key>
That reduces the chance of using the wrong authentication header.
7. How to verify the setup
After configuration, start Claude Code in your project directory:
claude
Ask for a simple project-level task:
Please read the current project structure and summarize the main modules.
If it responds normally, the basic setup is working.
If it fails, check:
- Whether
ANTHROPIC_BASE_URL is exactly https://api.deepseek.com/anthropic
- Whether
ANTHROPIC_AUTH_TOKEN contains a valid DeepSeek API key
- Whether the model names are spelled correctly
- Whether the account has enough balance
- Whether the current terminal really has the environment variables loaded
On macOS / Linux / WSL:
echo $ANTHROPIC_BASE_URL
echo $ANTHROPIC_MODEL
On Windows PowerShell:
echo $env:ANTHROPIC_BASE_URL
echo $env:ANTHROPIC_MODEL
8. Compatibility notes
DeepSeek's Anthropic-compatible API is not identical to the native Anthropic API.
According to DeepSeek's documentation, the compatible endpoint supports text, tool calls, streaming output, thinking, and related features. Some fields may be unsupported or ignored. Multimodal image input, document input, and some MCP-related message types may not be fully compatible.
The practical takeaway:
This setup is well suited to text-heavy and code-heavy Claude Code workflows, but it may not cover every advanced multimodal or specialized toolchain scenario.
If you mainly use Claude Code for code reading, refactoring, bug fixes, and test generation, the setup is often useful. If you rely on native Claude-only capabilities, test the workflow before making it your default.
9. Who this setup is for
This configuration is useful if you:
- Use Claude Code frequently and want lower API costs
- Mostly run coding tasks rather than multimodal tasks
- Work with larger project contexts
- Want to keep Claude Code's terminal workflow
- Want flexibility across model providers
In short, if you like how Claude Code works but want to try DeepSeek's cost and context advantages, this setup is worth testing.
10. Summary
Running Claude Code CLI with the DeepSeek API comes down to three steps.
First, set the API endpoint:
https://api.deepseek.com/anthropic
Second, put your DeepSeek API key in:
ANTHROPIC_AUTH_TOKEN
Third, configure the models:
deepseek-v4-pro[1m]
deepseek-v4-flash
This keeps the Claude Code CLI workflow while routing model calls through DeepSeek.
It is not a universal native-Claude replacement. It is a practical compatibility setup. Whether it fits depends on your task type, reliability needs, and cost budget.
References
