What is ccglass?

ccglass is a developer tool that lets you see the exact requests your AI coding assistant sends to the AI model behind the scenes. It works with Claude Code, Codex, OpenCode, DeepSeek-TUI, and Kimi. Install it globally via npm, run one command, and it sets up a local proxy that sits between your coding tool and the AI service it talks to. A web dashboard opens in your browser showing every request in real time. The problem it solves is specific: these coding tools are built in ways that bypass the standard proxy settings you might use to inspect network traffic with other tools. ccglass takes a different approach. It redirects the tool to send requests to a local address first, then forwards them to the real API. Because the local hop is plain unencrypted HTTP, there is nothing to intercept at the TLS level. No certificate setup is needed. The dashboard shows the full system prompt, every message in the conversation history, the list of tools the model was given, token counts, cache usage, and an estimated cost per request in US dollars. A conversation flow view shows which tools the model chose to use and how results were fed back. A turn-to-turn diff view lets you compare two consecutive requests to see exactly what changed. You can also export any captured request as a readable text transcript, Markdown, JSON, or a HAR file (a standard format for web traffic recordings). When wrapping Claude Code specifically, ccglass can register its own query tools via MCP (a protocol for giving AI models access to external tools), so the AI assistant can look up the requests it just sent during the same session. All captured requests are saved to a local folder called .ccglass in your project directory. Authentication tokens in request headers are masked by default to avoid accidentally committing secrets. The tool requires Node.js 18 or newer and has no runtime dependencies for its core features. It is MIT-licensed.