Free Claude Code: NVIDIA setup and MCP configuration

This walkthrough configures Free Claude Code, a local proxy that lets the Claude Code harness send requests to other model providers. The presenter uses installation instructions through Claude Code, opens the configuration window, creates an NVIDIA API key and applies it. He then adds an OpenRouter key and discusses keeping multiple providers available. The demonstrated inference runs through external APIs; running the proxy locally does not make the model requests offline.

After launching the FCC command in a terminal, he sends a short test prompt. The response interface displays an Anthropic model name and a cost estimate, but he explains that the actual backend is NVIDIA Nemotron. Those displayed labels should not be treated as proof of the model used or the provider's billing. He describes automatic fallback when a provider reaches a limit, without demonstrating a complete limit-exhaustion test.

The MCP section creates a separate Zapier server token, copies its URL into the configuration and discusses restricting its permissions. The closing advice distinguishes routine work from tasks needing stronger models and recommends testing outputs. The opening promise of unlimited access is not established by this setup: availability, rate limits and model quality vary across providers.