Why free-claude-code Is the Must‑Have Open‑Source Proxy for Claude Code

Why Claude Code Is Hot

Claude Code is Anthropic’s terminal‑level AI coding assistant that can write code, understand whole codebases, perform Git operations, and execute multi‑file refactoring via natural‑language commands. Its popularity has created three pain points: high API cost, difficult access from China, and model lock‑in.

free‑claude‑code Emerges

free‑claude‑code is a lightweight Python/FastAPI proxy that intercepts Claude Code’s requests, forwards them to free or low‑cost LLM providers (NVIDIA NIM, OpenRouter, DeepSeek, LM Studio, llama.cpp, Ollama), and translates the responses back into Claude‑compatible format, so Claude Code operates unchanged.

┌─────────────────┐   ┌──────────────────────┐   ┌──────────────────┐
│  Claude Code    │──▶│  Free Claude Code     │──▶│  Real LLM Vendor │
│  CLI / VSCode  │◀──│  Local Proxy (8082)   │◀──│  NIM / OR / Local│
└─────────────────┘   └──────────────────────┘   └──────────────────┘

Anthropic API format                OpenAI‑compatible format (SSE)

Supported Providers

NVIDIA NIM – free, 40 req/min, high‑quality models for daily use.

OpenRouter – free/paid, rate varies by model, many models for fallback.

DeepSeek – pay‑as‑you‑go, rate depends on plan, Anthropic‑compatible.

LM Studio – free local, unlimited, privacy‑focused.

llama.cpp – free local, unlimited, lightweight inference.

Ollama – free local, unlimited, easiest local solution.

Key Features

Zero‑intrusion: only two environment variables, no changes to Claude Code CLI or VSCode extension.

Thinking‑chain preservation: automatically parses <thinking> and reasoning_content tags for models like DeepSeek R1, GLM.

Tool‑call fixing: heuristic parser corrects malformed tool calls from open‑source models.

Smart throttling: built‑in concurrency control and 429 back‑off to avoid rate‑limit bans.

10‑Minute Quick Start

Visit build.nvidia.com, register a free account, complete phone verification, and copy the API key (e.g., nvapi‑xxx).

Clone the repository and install dependencies:

# Install uv (Python package manager)
curl -LsSf https://astral.sh/uv/install.sh | sh
uv python install 3.14

# Clone project
git clone https://github.com/Alishahryar1/free-claude-code.git
cd free-claude-code
cp .env.example .env

Edit .env to set the NVIDIA NIM key and model mapping:

NVIDIA_NIM_API_KEY="nvapi‑yourKey"
MODEL_OPUS="nvidia_nim/z-ai/glm-5.1"
MODEL_SONNET="nvidia_nim/z-ai/glm-5.1"
MODEL_HAIKU="nvidia_nim/z-ai/glm-5.1"
MODEL="nvidia_nim/z-ai/glm-5.1"
ENABLE_MODEL_THINKING=true

Start the proxy:

uv run uvicorn server:app --host 0.0.0.0 --port 8082

Configure Claude Code to point to the local proxy by adding to settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:8082",
    "ANTHROPIC_AUTH_TOKEN": "freecc"
  }
}

Run claude and verify a simple “Hello” query returns a response.

Advanced Usage

The project also offers a Discord/Telegram bot, multi‑model switching, and an interactive model selector claude‑pick that works with fzf for on‑the‑fly model choice.

Why Developers Star the Project

Higher success rate than earlier tools like CC Switch, thanks to unified model naming and clean request conversion.

Comprehensive documentation covering every .env field and provider integration.

Active maintenance with frequent provider additions (e.g., native Ollama Anthropic support).

MIT license enables free modification and downstream development.

Some users note that substituted models may not match Claude’s native capabilities for very long‑context, complex tool calls, or deep refactoring, but they handle everyday coding, single‑file edits, and code explanation well.

Conclusion

free‑claude‑code exemplifies a clever “proxy‑layer” approach: it does not attempt to replicate Claude Code but leverages its open API to let any LLM drive the powerful coding assistant, addressing cost, accessibility, and model‑lock concerns that dominate the AI‑coding ecosystem today.