Claude Code Router plugs free models into Claude Code. But the ccr code command every guide online gives you no longer opens Claude Code

CCR is a model gateway. It gives Claude Code, Codex, OpenCode and a few other agents one local address, and you manage the providers, models and keys behind it from a single dashboard. If a provider fails it can move on to the next model, and its logs show which model each request went to and how many tokens it used.

What is extra below: what replaced the old command, why you have to install Claude Code separately, the order to go through the dashboard in, why Ollama is not in the ready made list, and where the free tier actually ends.

Unlike how the videos put it, there is no trick fooling Claude Code. Claude Code already lets you change where it sends requests, and CCR sits at that address. The result is the same though: the interface is Claude Code, the model answering is not Claude. Set your quality expectations accordingly. I went through this distinction in detail at number 110 in this archive, using another tool built on the same idea.

You need two things. First, Node 22 or newer; the package itself requires it. Second, Claude Code itself. CCR does not install it, it only routes it. If Claude Code is not on your machine, the profile will not open.

node -v
# should be v22 or newer

# Claude Code, macOS / Linux:
curl -fsSL https://claude.ai/install.sh | bash

# Claude Code, Windows PowerShell:
irm https://claude.ai/install.ps1 | iex

npm install -g @musistudio/claude-code-router
ccr ui

The second command starts a background service and opens the dashboard in your browser. The dashboard is on port 3458, and the address Claude Code connects to is on 3456. If the install finished but the ccr command is not found, npm's global folder is not on your PATH: find it with npm prefix -g, add it to PATH and open a new terminal.

If you would rather skip the terminal, there is a desktop app for the same thing: download it for Windows, macOS or Linux from the Releases page of the repo. Same dashboard, same steps. The only difference is the launch command: the commands the app copies start with ccr-app, not ccr.

The video's three routes connect three different ways in the dashboard. NVIDIA and OpenRouter are in the ready made list and fill in the address themselves. Ollama is not in the list; you choose Other and type the address yourself.

When creating the profile, leave the scope on the default, Only opened from CCR. That way only the Claude Code you open through CCR goes to the free model, and your regular claude command keeps working with your subscription exactly as before. If you choose System default, every Claude Code session on your machine goes through CCR.

# use whatever you named the profile
ccr "free"

# once Claude Code is open, to switch between models:
/model

You can tell it is working from the Logs page in the dashboard: it shows which provider and model your message went to and how many tokens it used. To record anything, first turn on Request logs under Logs & Observability in Settings. If it is on and no entry shows up, the request did not go through CCR. Logs are kept for the current day only and cleared the next day.

The tool is free, that part is true. Unlimited is not, because the limit is set by the provider you connect, not by CCR. OpenRouter's free tier is 50 requests a day, and a coding agent fires dozens of requests for a single task; half an hour of serious work uses it up. Load $10 of credit once and the daily cap goes up to 1,000. NVIDIA gives you a free key, but you will see the credit allowance on your own account page. The only truly unlimited route is Ollama, and its price is hardware and model quality.

If you want to know how good a given free model really is, at number 113 in this archive I covered NVIDIA's Nemotron 3 Ultra in detail: where it stands against Claude on independent measurements, and what its free tier gives you. CCR's NVIDIA preset ships with this model by default too.