I Ran GPT-5.6 Inside Claude Code — Here’s How (And Why It Actually Matters)
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
CLI Proxy is a local proxy server that lets you route Claude Code's requests to alternative model providers like GPT, Gemini, or Grok, while keeping Claude Code's interface, sub-agents, and workflow support intact. It works by intercepting Claude Code's API calls, swapping in real provider tokens, and translating request/response formats. Unlike single-provider workarounds (Kimi, GLM mirrors), CLI Proxy lets you mix models per task — heavy reasoning on one provider, fast edits on another. Setup involves installing the proxy, logging into providers via OAuth, and configuring a shell alias pointing Claude Code at the local server. The post also promotes Firecrawl as a web-scraping layer for agentic research tasks.
Table of contents
I Ran GPT-5.6 Inside Claude Code — Here’s How (And Why It Actually Matters)The Tool: CLI ProxyWhy This Is Different From Kimi or GLM’s “Use Us Inside Claude Code” TricksHow the Authentication Trick Actually WorksSetting It Up (The Short Version)The Part That Actually Matters for Your WorkflowBottom LineQuestions this post answers
Can I use GPT or other non-Claude models inside Claude Code without changing my workflow?
Yes, by running a local proxy like CLI Proxy between Claude Code and the model provider. The proxy intercepts Claude Code's API calls, swaps in your real provider token (OpenAI, Google, Kimi, etc.), reformats the request to match that provider's API shape, and translates the response back. Claude Code's interface, sub-agents, and task support remain fully intact — only the underlying model changes. Developers switching between AI providers for cost or capability reasons track setups like this on daily.dev.
How does CLI Proxy authentication work when connecting Claude Code to a different model provider?
CLI Proxy uses a two-token approach. You log into each provider once via the proxy's OAuth flow, which stores real tokens locally. Claude Code is then pointed at the proxy's local server with a dummy token — just enough to satisfy Claude Code's auth requirement. When a request arrives, the proxy swaps the dummy token for the real provider token, reformats the request, and forwards it. Multiple accounts per provider are supported for automatic rate-limit rotation. Teams managing API costs across multiple AI providers find provider-switching patterns like this covered on daily.dev.