110September 5, 2026·1 items

What's free isn't Claude, it's the Claude Code interface. That distinction matters

The tool is real, open source, MIT licensed and was updated yesterday. But the name misleads. What's free isn't Anthropic's models, it's the Claude Code interface that runs in your terminal. FCC stands up a local proxy and routes requests to NVIDIA NIM, OpenRouter, or Ollama on your own machine. So you're using Claude Code, not Claude. Knowing that matters, because it sets your quality expectations, your account safety, and where free actually stops. The install isn't two commands and doesn't take five seconds either: there's an admin panel in the browser in the middle of it. Below: the real flow, the real limits, and the forty seven other providers the video never mentions.

Claude CodeÜcretsiz APIGeliştirici Araçları

The tool itself

The repo is here. Open source, MIT licensed, written in Python. It opened in January 2026, has 53,241 stars and 8,547 forks, and was last updated on 5 September 2026. So it isn't an abandoned project.

What's extra below: what this tool actually does, the real install steps, why your account doesn't get banned, exactly where the free tier stops, and the forty seven providers and nine coding agents the video never mentions. Plus one easily missed trap: if you set up a fallback model list, a single request can eat quota from more than one provider.

Open the repo

What's free is not Claude

This is the most important sentence on the page. Claude Code is two separate things: the interface that runs in your terminal, and the model behind it. FCC leaves the interface exactly as it is and only changes where the model comes from. It opens a local proxy on your machine, Claude Code talks to that, and the proxy forwards the request to whichever provider you picked.

What runs on top of what

Claude Code

The interface you see in the terminal. Unchanged, stays as it is

FCC (local proxy + admin panel)

Runs on your machine, catches the request and routes it

The provider

NVIDIA NIM, OpenRouter, Ollama and forty seven more

The model

Nemotron by default. DeepSeek, Kimi, GLM and Qwen are all selectable

The conclusion: you're not getting free access to Anthropic's Opus or Sonnet, they aren't in this chain at all. The default model is NVIDIA's Nemotron. The video says "models as powerful as Claude behind the scenes"; that isn't a verifiable claim, and the right expectation is this: you get good open models, not Claude itself. The tool itself doesn't make that claim either.

The install isn't two commands and doesn't take five seconds

The video says "you run these two commands on screen". That isn't the real flow, and the difference matters because the skipped step isn't even a command, it's a panel that opens in your browser. That's exactly where people who stare at the terminal afterwards wondering why nothing works have gone wrong.

The actual order

1
The install command
A single command. When prompted you pick at least one coding agent
2
Start FCC
On Windows and macOS open the desktop app, on Linux run fcc-server
3
Configure in the admin panel
This is the skipped step. It opens in the browser: paste the API key, pick a model, hit Apply
4
Run the agent
fcc-claude. There are separate commands for Codex, Pi, OpenCode and Aider

The install script downloads more than you'd think

I read the script: it installs Python 3.14 and uv, and on top of that it installs Claude Code, Codex, Pi, OpenCode, Hermes, DeepSeek Harness, Grok Build, Muse Code and Aider by default. Cline is off by default. You can deselect what you don't want when prompted, and you should. That's also why this isn't a five second job. If you want out, the uninstall script removes FCC and the .fcc folder but leaves Python, uv and the installed agents behind.

Will your account get banned

That's the first question the phrase "free Claude Code" raises, and it's a fair one. The answer is no, because nobody is being tricked here. Anthropic's systems aren't being accessed, no account is shared, no limit is circumvented. An open source terminal tool is simply connecting to free tiers that other companies offer on purpose. You sign in to each provider with your own account and your own key.

The project takes a clear position on this: it describes its providers as terms-of-service friendly and says it removes an integration if it stops being allowed. That isn't just talk; when GitHub Models shut down in July they noted it in the README and added behaviour that resets selected models back to your default. Some providers' plans are also for personal, interactive use only, and they spell that out separately for Kimi Code and the QwenCloud Coding Plan.

Where free stops, exactly

This is the part the video skips and the part that actually decides everything. The free tier isn't offered by the tool, it's offered by the providers, and they set the limits. The project says so itself: free tier availability and limits are controlled by each provider and can change.

The real limit on the video's three methods

Method	What you get for free	Where you hit the wall
OpenRouter	20 requests a minute, 50 a day	50 a day is very little for a coding agent. Buy $10 of credit once and the daily cap jumps to 1,000
NVIDIA NIM	A free key and a wide model catalog	Third party sources disagree on the credit allowance, so I'm not quoting a number. You'll see it on your own account page
Ollama	No limit, everything runs on your own machine	The cost is hardware and model quality. See the section below

The OpenRouter number deserves a pause, because that's where "completely free" strains hardest. 50 requests a day won't survive half an hour of serious coding; an agent fires dozens of requests for a single task. Opening extra accounts doesn't help either, since they govern capacity globally rather than per account. So on the OpenRouter side, real usage means leaving $10 there once.

A fallback list can double-spend your quota

In the admin panel you can set an ordered fallback model list; if one provider goes down it moves to the next and your session isn't interrupted. Sounds good, but the project's own warning is this: a failed request may reach and consume usage from more than one provider before it succeeds. When your allowance is 50 requests a day, that's a serious matter. Put your most generous provider first when you build the list.

The video says three methods, the catalog has fifty providers

NVIDIA NIM, OpenRouter and Ollama are just the quick start examples. The README table lists close to fifty providers, and some of them are things you already pay for: your ChatGPT subscription, your GitHub Copilot subscription, Google AI Studio, Groq, Cerebras, Together, DeepSeek, Mistral, Z.ai. So this isn't only a "find free models" tool, it's a way to wire every model you already have access to into Claude Code from one place.

The same goes for the agent side. The video only mentions Claude Code, but you can use the same model pool with Codex, Pi, OpenCode, Cline, Hermes, DeepSeek Harness, Grok Build, Muse Code and Aider. Each has its own launch command: fcc-codex, fcc-pi, fcc-opencode and so on. That's useful when you like an agent's interface but not its model.

Running locally: the graphics card isn't the point

The video says "if you have a strong graphics card you can run it locally with Ollama". A graphics card is necessary but not sufficient, and the real obstacle is elsewhere. The project's own warning: prefer tool-capable models for coding agents, and local models need enough context to hold the agent's system prompt and tool definitions.

My graphics card is strong, I'll download any model and wire it into Claude Code.

The model has to support tool calling. A coding agent calls tools to read and write files and run commands; a model without it will chat but won't do the work.

A small model will do, my questions are short anyway.

Your question may be short but the agent's system prompt and tool definitions are not. The model needs enough context to carry those before you type anything at all.

Ollama isn't the only local option either: LM Studio and llama.cpp are supported too, and cloud-hosted Ollama Cloud sits there as a separate provider. Ollama setup is two commands, then you enter the model tag with the ollama/ prefix in the admin panel.

free-claude-code (FCC)

An open source local proxy that connects Claude Code and nine other coding agents to the free or already-paid tiers of nearly fifty providers. The interface doesn't change, only the model behind it does. MIT licensed, written in Python, actively developed.

To update you run the same install command again, there's no separate command
The uninstall script is separate: uninstall.sh and uninstall.ps1, at the same address
For other agents there are fcc-codex, fcc-pi, fcc-opencode, fcc-aider and similar commands
If you'd rather not use the fcc- launchers you can wire it up by hand: the proxy sits on localhost:8082, so pointing ANTHROPIC_BASE_URL at it is enough. That's also how you connect VS Code and JetBrains

Installation

# 1. Install. macOS / Linux:
curl -fsSL "https://raw.githubusercontent.com/Alishahryar1/free-claude-code/main/scripts/install.sh" | sh

# Windows PowerShell:
& ([scriptblock]::Create((irm "https://raw.githubusercontent.com/Alishahryar1/free-claude-code/main/scripts/install.ps1")))

# 2. Start FCC. On Windows and macOS open the desktop app.
# On Linux:
fcc-server

# 3. In the admin panel that opens, enter your API key, pick a model, hit Apply.

# 4. Run Claude Code:
fcc-claude

The install script sets up Python 3.14 and uv, then downloads nine coding agents by default; deselect what you don't want when prompted. If you'd rather read the script before running it, it sits in the scripts folder in the repo and the project itself suggests doing that. The step not to skip is the third: the admin panel opens in your browser and that's where the key goes, not the terminal.

Open

DOA: Yapay Zeka ve Otomasyon

Installing these tools on your own is one thing; actually building with them is another. The community has people using these daily and people building systems from scratch.

If you want a system that actually runs in your business, let's talk for 10 minutes; I'll look at what you're trying to build and tell you which path fits. Free, and not a sales pitch.

Book a 10-minute call

If you'd rather learn this alongside people doing the same work instead of on your own, the community is always open:

Join the community

This is an affiliate link.

← All resources