Who free-claude-code is for#
Individual developers reducing Claude Code API spend
FCC's primary documented use case is giving developers who already use Claude Code a way to continue using the same interface and agent against free provider quotas. The fcc-claude launcher starts Claude Code pointed at the local proxy; from there the workflow is identical to the native Claude Code experience.
Skip if:
Skip if you want a managed service with no configuration. FCC requires provider setup and tolerance for free-tier changes. Developers with consistent, heavy API usage may find a direct Anthropic subscription more predictable.
Teams running multiple coding agents across workloads
FCC's nine-agent catalog lets a team standardize on one proxy while different members use different agents. Claude Code, Codex, and Cline can each run against different models from the same FCC instance, with independent fallback chains per agent.
Skip if:
Skip if your team needs enterprise seat management, audit logs, and policy controls. FCC does not provide those. GitHub Copilot for Business or Amazon Q Developer include team management features that FCC lacks.
Developers experimenting with local and alternative models
FCC supports local models through Ollama, LM Studio, and llama.cpp alongside cloud providers. A developer testing which model performs best on their codebase can switch the active model in the Admin UI without changing any agent configuration.
Skip if:
Skip if your interest is local models only. You can connect Claude Code directly to a local Ollama server without the FCC proxy layer.
The problem it solves#
AI coding assistants like GitHub Copilot, Claude Code, and Amazon Q Developer charge per seat or per token. A developer using Claude Code for full sessions can exhaust a free-tier quota in hours; the next tier is a monthly subscription that compounds across team members. Provider outages add a second problem: when one API goes down, the coding session stops until the developer switches models manually, often losing context.
The output token problem is less obvious but equally costly. Coding agents generate substantial terminal output for every command they run, and that output counts against API quotas even when it carries no useful information. Teams that want to run multiple agents across different workloads, or route different model tiers to different tasks, have no built-in cross-provider orchestration from any commercial provider.
How it solves it#
50+ provider catalog with automatic fallback
FCC connects to more than 50 providers including NVIDIA NIM, OpenRouter, Groq, xAI, Google AI Studio, and local models via Ollama or LM Studio. When a provider fails after retries, FCC automatically routes the request to the next configured model without requiring you to restart the coding agent or lose session context.
Nine coding agents from one model catalog
Run Claude Code, Codex, Pi, OpenCode, Cline, Hermes, DeepSeek Harness, Grok Build, or Muse Code against any FCC-configured provider. A unified Admin UI sets the active model and fallback list, and individual Claude Code tiers (Fable, Opus, Sonnet, Haiku) can each be routed to a different provider model independently.
Up to 90% fewer terminal output tokens with RTK
The optional RTK filter strips common command output before it reaches the provider, reducing output token consumption by up to 90%. Five additional FCC-native optimizations handle quota probes, command-prefix detection, title generation, suggestions, and filepath resolution without any provider call.
Voice input via local Whisper or NVIDIA NIM
FCC supports voice-to-code workflows using local Whisper transcription or NVIDIA NIM ASR, so you can speak commands and have them transcribed locally before being sent to the coding agent. No audio leaves the machine when using local Whisper.
Works in terminal, VS Code, JetBrains, and phone
Coding agents connect to FCC through native launchers (fcc-claude, fcc-codex, and others) or through editor integrations for VS Code and JetBrains by pointing the extension's base URL to the local FCC server. Discord and Telegram integrations are also available for phone-based use.
Strengths and trade-offs#
Strengths
- MIT licensed with 1.3B+ free tokens monthlyFCC is MIT licensed with no usage restrictions. The 1.3B+ monthly token figure aggregates free tiers across more than 50 providers, so individual provider limits do not cap total availability the way a single commercial subscription would. Providers that change their terms are removed from the catalog.
- ToS-friendly provider selection by designFCC explicitly documents provider terms and removes integrations that stop being allowed. This is a direct contrast to approaches that bypass rate limits through unofficial means. The README states that FCC follows provider terms and updates the catalog when policies change.
- Local proxy with no intermediary data routingFCC runs on localhost and forwards requests directly to whichever provider you configure. Unlike commercial AI coding tools that route all traffic through their own infrastructure, FCC adds no intermediary between your code and the provider endpoint.
- Per-tier model routing for Claude Code usersUsers of Claude Code can route each model tier (Fable, Opus, Sonnet, Haiku) to a different provider independently. This lets you assign cheaper or faster models to lightweight tasks and reserve higher-capability models for complex sessions, without changing your Claude Code workflow.
Trade-offs
- -Provider free-tier limits change without noticeFree-tier token allocations are set by each provider and can change at any time. The 1.3B+ total is an aggregate across all 50+ providers; any individual provider may reduce, suspend, or remove its free tier. The README explicitly notes this: free-tier availability and limits are controlled by each provider and may change.
- -Requires provider accounts and manual configurationInstalling FCC takes a single shell command, but configuring useful providers requires creating accounts at NVIDIA NIM, OpenRouter, Groq, or others, obtaining API keys, and entering them in the Admin UI. Users who want a turnkey coding assistant with no setup will find this more involved than a commercial product like GitHub Copilot or Amazon Q Developer.
- -Young project with 356 open issues at time of writingThe repository reached its first public release in January 2026 and shows 356 open GitHub issues. The issue count reflects active user adoption alongside ongoing stability work. Users running production coding workflows on FCC should monitor the issue tracker for regressions affecting their provider or agent combination.
free-claude-code vs alternatives#
Free Claude Code vs GitHub Copilot
GitHub Copilot is a subscription-based coding assistant integrated into VS Code, JetBrains, and other editors at $10-19 per user per month. FCC is a local proxy that lets you run the Claude Code and Codex agents against free provider tiers, serving a different deployment model entirely.
| Feature | Free Claude Code | GitHub Copilot |
|---|---|---|
| License | MIT | Proprietary |
| Cost | Free (provider tiers) | $10-19/user/month |
| Self-hosting | Yes, runs on localhost | No |
| Coding agents supported | Claude Code, Codex, 7 others | Copilot Chat only |
| Provider choice | 50+ | Single (GitHub/OpenAI) |
FCC is the better choice when you want to use the actual Claude Code or Codex agents against free provider quotas rather than GitHub's built-in assistant, or when you want a self-hosted setup with no per-seat fees. GitHub Copilot is the better choice for teams that need enterprise seat management, audit logging, and policy controls, none of which FCC provides. Copilot also has no per-user configuration overhead; FCC requires provider account setup.
Free Claude Code vs Anthropic (Claude)
The Anthropic API is what the Claude Code agent uses by default when you pay directly. FCC intercepts those requests and routes them to third-party providers that offer compatible APIs at no direct cost.
| Feature | Free Claude Code | Anthropic Claude API |
|---|---|---|
| License | MIT | Proprietary |
| Cost | Free (provider tiers) | Pay per token |
| Self-hosting | Yes, runs on localhost | API only, no self-host |
| Model access | 50+ providers (Claude-compatible) | Anthropic models only |
| Support | Community | Official |
FCC is the better choice when you want to continue using Claude Code without Anthropic's per-token rates and when a compatible open model from NVIDIA NIM, Groq, or another provider meets your quality bar. The Anthropic API is the better choice when you specifically need Anthropic's proprietary Claude models at full capability: FCC routes to compatible models from other providers, not to Anthropic itself. For tasks where only a genuine Claude model will do, the direct API is the more reliable path.
Install and self-host#
Install with the shell script; on macOS and Windows, use the desktop app after install.
```bash
curl -fsSL "https://raw.githubusercontent.com/Alishahryar1/free-claude-code/main/scripts/install.sh" | sh
```What it's built on#
- Languages
- JavaScriptPython
- Frameworks
- FastAPI
FAQ#
Is Free Claude Code affiliated with Anthropic?
No. Free Claude Code is an independent open source project. The README explicitly states it is not affiliated with or endorsed by Anthropic, and that Claude and Claude Code are Anthropic trademarks. FCC works by routing the Claude Code agent's API calls to third-party providers that offer compatible APIs.
Does using Free Claude Code violate any terms of service?
FCC routes requests to other providers rather than bypassing Anthropic directly. Whether a specific provider permits the use case depends on that provider's policies. FCC documents provider terms and removes integrations that are no longer allowed, which is what the ToS-friendly label refers to in the README.
How do I install Free Claude Code?
On macOS or Linux, run the install script via curl: curl -fsSL https://raw.githubusercontent.com/Alishahryar1/free-claude-code/main/scripts/install.sh | sh. On Windows, use the PowerShell installer listed in the README. After installation, start the server with fcc-server on Linux or via the desktop app on macOS and Windows, then run fcc-claude or another agent launcher.
What providers offer free tokens through Free Claude Code?
The catalog includes more than 50 providers, including NVIDIA NIM, OpenRouter, Groq, Google AI Studio, xAI, QwenCloud, GitHub Models, and local providers like Ollama and LM Studio. Free-tier availability and token limits are controlled by each provider and may change at any time without notice.
Can I use Free Claude Code in VS Code?
Yes. Install the Claude Code extension from the VS Code Marketplace, then add FCC's local server URL and auth token to the extension's environment variables in your VS Code user settings JSON. The full configuration snippet is in the FCC README under Connect Your Client.
Similar open-source tools#
codex
OpenAI's terminal coding agent, Apache-2.0 licensed
compound-engineering-plugin
Add CE planning and review workflows to agents
garden-skills
Production-ready agent skills for Claude Code, Cursor, and Codex
fx
Native Zig CLI coding agent, 7.8 MiB, any AI model.
kilocode
Open source AI coding agent. 500+ models at zero markup.
prime-agent
Automate coding workflows with a self-improving agent