An MIT-licensed open source project that lets you run Claude Code without an Anthropic API key. A local server (proxy) you start on your own machine receives requests from Claude Code and forwards them to inference providers with free tiers such as NVIDIA NIM, Groq, and OpenRouter, or to local models running on LM Studio, llama.cpp, or Ollama. It supported four providers when it first appeared, but as of August 2026 it covers roughly 50 providers and nine coding agents including Claude Code, Codex, and Cline, and has gathered more than 50,000 stars on GitHub.
Key Features
- Start Claude Code without an API key: Launch
fcc-server, then runfcc-claude, and Claude Code calls models through this tool. No paid Anthropic account is required - Around 50 providers behind one interface: NVIDIA NIM, Groq, OpenRouter, Together AI, Cerebras, Google AI Studio, Cloudflare Workers AI and others can be used across free-tier, pay-as-you-go, and subscription plans. The README claims “1.3B+ free tokens every month”
- Local model support: Connects to LM Studio (localhost:1234 by default), llama.cpp (8080), and Ollama (11434), so you can build a setup that never sends code outside your machine
- Automatic fallback: When a provider goes down or hits its limit, it switches to the next model without a restart
- Nine supported agents: Besides
fcc-claude(Claude Code), there arefcc-codex,fcc-cline,fcc-opencode,fcc-piand more ─ change the command and reuse the same configuration across agents - Admin UI plus Discord / Telegram bots: A browser admin screen configures API keys, the model in use, the fallback order, and reasoning depth. The Discord and Telegram bots offer
/stats,/stop, and/clearso you can check status and stop work from your phone, with voice input supported
Pricing
| Item | Cost | Details |
|---|---|---|
| Free Claude Code itself | Free (MIT license) | Self-hosted. Install from GitHub and run it on your own PC |
| Providers with free tiers | $0 (within each provider’s free allowance) | Free models on NVIDIA NIM, Groq, OpenRouter and others. Limits follow each provider’s rules |
| Paid providers | Each provider’s usage-based pricing | When you connect OpenAI, Google Vertex AI, Amazon Bedrock and similar services |
| Local models | $0 (excluding electricity and hardware) | Via LM Studio / llama.cpp / Ollama |
Pricing is as of August 2026. The contents and limits of free tiers change at each provider’s discretion. Check the official repository and each provider’s pricing page for the latest information.
Pros & Cons
✅ Pros
- Try the Claude Code workflow without an Anthropic subscription or API billing
- Switch between free tiers, paid APIs, and local models from a single admin screen, which keeps switching costs low
- Automatic fallback means work is less likely to stop when you hit a free-tier limit
- MIT-licensed open source, so you can inspect and modify what is sent where
- Discord / Telegram integration lets you check on the agent and stop it while away from your desk
⚠️ Cons
- An independent project, not an official Anthropic product. Claude and Claude Code are Anthropic trademarks, and official support is not available
- Output quality depends entirely on the model behind it. This does not give you Claude Opus or Sonnet themselves for free
- Running on local models requires enough context length to hold the agent’s long system prompt, plus tool-calling support
- Initial setup takes effort: Python 3.14 and uv, plus obtaining an API key for each provider
- Free tiers can change or end without notice, so it is not suited to production business use
Comparison with Similar Services
| Item | Free Claude Code | Claude Code Router | LiteLLM | OpenRouter |
|---|---|---|---|---|
| Form | Self-hosted proxy (OSS) | Self-hosted router (OSS) | Self-hosted proxy (OSS) plus a cloud version | Hosted API gateway |
| Main aim | Run various agents on free tiers and local models | Route Claude Code requests to other models | Unify many LLM APIs behind an OpenAI-compatible interface | Use many models with a single key |
| Supported agents | Nine (Claude Code / Codex / Cline and others) | Mainly Claude Code | Agent-agnostic | Agent-agnostic |
| Cost | Free (provider costs are separate) | Free (provider costs are separate) | OSS version free, cloud version paid | Usage-based (some free models) |
| Notable | Admin UI, automatic fallback, chatbot integration | Lightweight routing configuration | Operational features such as monitoring and cost management | Easy to adopt, one key is all you need |
Who Is It For
- People who want to try the Claude Code workflow once before paying for it
- People who want to combine free-tier inference services to keep learning and personal projects cheap
- People who handle code that cannot leave the company and want to run a coding agent on local models
- People who want to compare several coding agents using the same configuration
- Developers who can handle setup and troubleshooting themselves
Summary
Free Claude Code is an open source tool that places your own proxy between a coding agent and model providers, letting you run Claude Code without an Anthropic API key. Roughly 50 providers and local models can be chosen from one admin screen, and automatic fallback is designed to absorb free-tier limits. It is not an official tool, and output quality depends on the model behind it, so the practical approach is to try it for learning and personal projects first, then decide on business use only after checking each provider’s terms and the quality you actually get.