My setup for running Moonshot's Kimi K3 with its 1M-token context as a model in Claude Code. A local teamclaude proxy holds the Kimi API key next to my Claude accounts and routes by model name. Claude Code needs no endpoint or token configuration at all, and the subscription limits show up in the UI like native ones.
The parts:
- Kimi for Coding: Moonshot's subscription with an Anthropic-compatible endpoint and the
k3model. - teamclaude: local multi-account proxy (
@karpeleslab/teamclaudeon npm, being renamed TeamRouter). - Claude Code: one line in settings.json.
- ailine: statusline that renders the quota bars.