jianshuo

jianshuo / ccglass

Public

See what your coding agent (Claude Code, Codex, Kimi) sends to the model — local proxy + web dashboard

14
0
85% credibility
Found May 23, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

ccglass is a local debugging tool that shows you exactly what your AI coding assistant (Claude Code, Codex, or Kimi) sends to the AI model. When you run it, it starts a small local dashboard and launches your chosen coding assistant. Every request and response then appears in real-time on the dashboard, letting you see the full system instructions, message history, tool usage, token counts, and costs. You can click into any request for details, compare two requests to see what changed, and export conversations for later. Everything stays on your computer — no data leaves except to the actual AI service you're using.

How It Works

1
💬 You hear about a way to see what your AI coding assistant is doing

A developer friend tells you about a tool that lets you peek inside your coding assistant's conversations with the AI model.

2
📦 You install it with one simple command

You type a quick install command and the tool is ready on your computer, no complex setup needed.

3
🎯 You choose which coding assistant to watch

When you run the tool, it asks you whether you want to watch Claude Code, Codex, or Kimi — pick the one you're using.

4
🪟 A dashboard opens and your coding assistant starts

A small window appears showing a live dashboard, and your chosen coding assistant launches with everything automatically connected.

5
👀 You watch every request appear in real-time

As you use your coding assistant, each question and answer shows up instantly on the dashboard — you see the full conversation happening.

6
You explore what's being sent to the AI
📊
See token counts and costs

Check exactly how many tokens each request used and what it cost in dollars.

🔄
Compare two requests side-by-side

Pick any two requests to see a side-by-side diff showing what was added or removed this turn.

💾
Save any conversation for later

Export any request to your computer as a readable document to share or archive.

✨ You now understand exactly what your AI assistant is doing

No more guessing — you can see every detail of how your coding assistant talks to the AI, helping you debug issues and understand the process.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is ccglass?

ccglass is a local reverse-proxy that intercepts API traffic from coding agents like Claude Code, Codex, and Kimi. It runs entirely on your machine, spinning up a dashboard where you watch every request in real-time: the full system prompt, message history, tool schemas, and token counts. The agent still calls the real API directly over HTTPS; ccglass only sees the plain HTTP hop to localhost, so no CA certificates or TLS manipulation required. It works today, and it keeps working across agent updates because it never touches the outbound TLS connection.

Why is it gaining traction?

The hook is simplicity: `ccglass claude` and you're watching. Unlike Charles or mitmproxy, this doesn't require installing a root certificate or fighting proxy settings. Unlike patching `fetch` or wrapping SDKs, this survives client updates. You get token breakdowns, cache hit rates, estimated costs per request, and a turn-to-turn diff that shows exactly what new context was added each turn. Developers are using it to debug prompts, understand context usage, and figure out why their agent keeps calling the same tool repeatedly.

Who should use this?

Prompt engineers and developers debugging Claude Code or Codex behavior will get the most value. If you've ever wondered "what exactly is this agent sending?" or tried to optimize context window usage, this is purpose-built for that. It's also useful for anyone monitoring costs on high-volume agent workflows.

Verdict

This is a sharp, focused tool with a solid implementation and no runtime dependencies. The zero-dependency constraint means it installs fast and breaks rarely. That said, 14 stars signals early days, so expect minor rough edges and a sparse README. The credibility score of 0.85% reflects this youth but the MIT license and clean code structure suggest a project worth watching. Try it if you want visibility into your agent's API calls; wait for maturity if you need battle-tested tooling.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.