I built a local reverse proxy to see what Claude Code actually sends to Anthropic

侯垒 — Wed, 10 Jun 2026 07:08:09 +0000

The problem I couldn't solve

I was spending ~$1,800/month on Claude Code.

I had no idea where the money was going. I had no idea which prompts were 4,000-token monstrosities, which ones were 200-token gems, or which ones I'd accidentally repeated 3 times this week.

I tried the obvious tools first:

mitmproxy — didn't work. Claude Code (and Codex, DeepSeek, Kimi, GLM, etc.) all ignore HTTP_PROXY because they're native CLIs that open HTTPS sockets directly.
Charles — same problem.
Langfuse / Helicone — these are SaaS. You have to send your data to them. Not what I wanted.
Custom hooks — limited to events the CLI exposes. I wanted the raw HTTP.

I wanted a local, open-source, zero-account way to see what my coding agent was doing.

The solution: a local reverse proxy on the loopback

The insight: every coding agent CLI talks to api.anthropic.com (or similar). If I make it talk to http://127.0.0.1:port instead, and have a tiny proxy on that port forward to the real API, the local hop is plain HTTP — easy to log, no CA cert, no TLS pinning pain.

That's it. That's the whole trick.

npm i -g ccglass
ccglass claude
# → opens http://localhost:8123 in your browser
# → real-time dashboard of every request

What I learned in 30 days

After running every Claude Code session through it, I found:

1. I had a 38% cache hit rate I didn't know about

I was repeating myself in 38% of prompts and paying full price. The dashboard made it visible. I rewrote my CLAUDE.md to front-load context — cache hit rate jumped to 70%, monthly bill dropped 35%.

2. Per-provider cost varies 10x

Same task:

Claude Sonnet 4.6: $0.42
GPT-4o: $0.31
DeepSeek: $0.04

I started picking per-task. Anthropic for quality, DeepSeek for bulk.

3. Turn counts were higher than I thought

Average 4.2 turns per task. After seeing the data, I rewrote my CLAUDE.md. Turn count dropped to 2.8. Less back-and-forth = less cost = faster delivery.

4. MCP self-inspection is wild

ccglass has an MCP server. When you run ccglass claude, the agent can query its own request history inside the chat. I asked Claude "what did I prompt you with 3 turns ago?" and it answered correctly.

What's supported

16+ providers out of the box:

Coding agents: Claude Code, Codex, OpenCode, CodeBuddy, Reasonix
Pure LLM APIs: Anthropic, OpenAI, DeepSeek, Kimi, GLM, OpenRouter
Cloud: AWS Bedrock, GCP Vertex AI
Local: Ollama, LM Studio

The limits (I want to be honest)

Cursor subscription models can't be intercepted (they use a server-side proxy).
VS Code Continue with built-in models: same.
It's local-only by design. No SaaS, no telemetry, no account. (If you want cloud, use Langfuse.)

Open source

GitHub: https://github.com/jianshuo/ccglass

460+ stars at time of writing. MIT licensed. PRs welcome.

If you ship with Claude Code / Codex / Kimi and have ever asked "where is my money going", try it once. The data is eye-opening.

Make Your AI Coding Agent Transparent - See What It Actually Sends to the Model

侯垒 — Wed, 10 Jun 2026 02:46:06 +0000

If you've been using AI coding agents like Claude Code or Codex, you know how powerful they can be. But they also feel like a black box. What's actually in that system prompt? How much context is being sent every turn? Where are all my tokens going?

I recently found a tool called ccglass that answers these questions beautifully, and I wanted to share my experience.

What is ccglass?
ccglass is a lightweight local tool that acts as a reverse proxy between your AI coding agent and the model API. It intercepts all requests and responses, logs them, and displays them in a really nice web dashboard.

ccglass Dashboard

Getting Started
Installation is simple (Node.js 18+ required):

npm install -g ccglass

Then just run it and pick your agent:

ccglass

Or specify directly:

ccglass claude # for Claude Code
ccglass codex # for Codex
ccglass deepseek # for DeepSeek-TUI

It will:

Start a local proxy server
Set the right environment variables automatically
Launch the agent for you
Open the web dashboard

That's it! No CA certificates to install, no complicated setup.

What You Can See

The dashboard shows you everything:

The Full System Prompt

This is probably the most interesting part. You get to see how different agents frame their instructions to the model. Claude Code's system
prompt is fascinating to read!

Complete Message History

See the full context being sent each turn, how it evolves, and what gets kept vs. dropped.

Tool Schemas and Calls

See what tool definitions the agent provides to the model, and what tool calls the model makes in response.

Token and Cost Breakdown

Know exactly how many tokens you're using, what's cached, and get cost estimates per request and per session.

Token Summary

Turn-by-Turn Diff View

Compare requests to see exactly what changed between turns. Super useful for debugging why an agent started behaving differently.

Supported Agents

The list is pretty extensive:

Claude Code (including Bedrock and Vertex modes)
Codex (OpenAI)
DeepSeek-TUI and Reasonix
Kimi (Moonshot)
OpenCode
Ollama
OpenRouter
GLM/Zhipu
CodeBuddy (VS Code/JetBrains plugins)

IDE Support

If you use Cursor, Cline, Continue.dev, or similar IDEs that let you set a custom API base URL, you can use the proxy mode:

ccglass proxy --provider openai

Then just point your IDE's API base URL to the local proxy address it gives you.

Why I Like This

Learn from the best - See how production-grade agent systems are built
Debug effectively - Understand why your agent is making certain choices
Optimize costs - See where your tokens are actually going
Local only - All logs stay on your machine (default redacts auth tokens)
MIT licensed - Completely open source

Pro Tip: Export for Documentation

You can export any request to raw HTTP, Markdown, JSON, or HAR format:

ccglass export / --format md

Great for documentation, bug reports, or just saving interesting prompts.

Try It Out

If you're using any AI coding agent regularly, I highly recommend giving ccglass a try. It will change how you think about these tools.

Install it now: