Building ccglass: the architecture of a local LLM reverse proxy

Short summary

ccglass is an open-source Node.js reverse proxy that captures LLM API traffic from coding agent CLIs (Claude Code, DeepSeek, Codex) by intercepting at the loopback level. Instead of attempting HTTPS interception, it overrides API base URLs to plain HTTP (127.0.0.1:8123), logs requests in real-time, and computes streaming costs incrementally. It includes a web dashboard, SSE streaming support, per-provider pricing models, and an MCP server for querying request history.

• •Local HTTP reverse proxy intercepts LLM CLI traffic without HTTPS certificates
• •Real-time dashboard shows prompts, costs with cache hits, and usage patterns
• •Built-in MCP server lets Claude Code query its own request history

Generated with AI, which can make mistakes.