Hey Product Hunt, I built ccglass because I wanted a simple way to see what AI coding agents are actually sending to the model. When using Claude Code, Codex, OpenCode, DeepSeek-TUI, Kimi, or other coding agents, a lot happens behind the scenes: system prompts, tool schemas, message history, tool calls, streamed responses, token usage, cache hits, and cost. Normal HTTP proxy tools often miss this traffic because these clients do not always respect HTTP_PROXY/HTTPS_PROXY, and TLS interception can be painful. ccglass is a lightweight local reverse proxy plus dashboard. Run one command, choose the agent you want to inspect, and it opens a local dashboard showing every request in real time. A few things it can show: - Full system prompts and message history - Tool schemas, tool calls, and tool results - Token usage, cache hit rate, latency, and estimated cost - Reassembled streamed responses - Turn-to-turn diffs - Export to raw HTTP, Markdown, JSON, or HAR It is open source, runs locally, and is designed for people building with AI coding agents who want more visibility into what is really happening. I would love feedback from developers using Claude Code, Codex, OpenCode, or similar tools. What should ccglass inspect next?

ccglass - Inspect what AI coding agents send to LLMs

Replies