cctrace

See what Claude really sends.

看看 Claude 到底发了什么。

LLM tracing — what Claude Code sends and receives each turn. Security & privacy tracing — what actually leaves your machine. One command.

LLM 追踪：Claude Code 每轮到底发送和接收了什么。 安全/隐私审计：哪些请求出站、去了哪些 host、payload 里带了什么。一行命令。

$ npm install -g @thevibeworks/cctrace

or bunx @thevibeworks/cctrace to run without installing 或 bunx @thevibeworks/cctrace 直接运行，无需安装

Captures /v1/messages, OAuth, usage/credits, MCP registry, bootstrap, and telemetry — not just the chat endpoint.

捕获 /v1/messages、OAuth、用量/额度、MCP registry、bootstrap、遥测 — 不只是聊天端点。

Filter chips with live counts, colored category badges, expandable headers and bodies, decoded SSE streams. System, light, and dark themes.

带实时计数的筛选标签、彩色分类徽章、可展开的 headers/bodies、解码后的 SSE 流。支持跟随系统、浅色和深色主题。

Every run saves a self-contained HTML file that renders the same UI offline. Send it to a colleague — no server needed.

每次运行都保存一份自包含 HTML，离线也能渲染同样的界面。发给同事就能看，不需要服务器。

Dynamic cert generation intercepts ALL hosts — Anthropic, Datadog telemetry, MCP servers. See every byte that leaves your machine.

动态证书生成拦截所有主机 — Anthropic、Datadog 遥测、MCP 服务器。看到离开你机器的每一个字节。

Credentials are redacted from headers, bodies, and URLs before anything hits disk. Your API keys stay your API keys.

凭据在落盘前就已从 headers、bodies 和 URL 中脱敏。你的 API key 还是你的 API key。

Auto-generates a CA, auto-detects your Claude install, and captures everything by default. Run cctrace and you're done.

自动生成 CA、自动识别 Claude 安装、默认捕获全部。运行 cctrace 就完事了。

Coming soon

Full interception 完整拦截 Dynamic certs for all hosts — Datadog, MCP servers, everything 为所有主机动态生成证书 — Datadog、MCP 服务器，全覆盖
Smart categories 智能分类 Count Tokens separated from Messages, External host filter Count Tokens 独立于 Messages，外部主机筛选
Copy + token summary 复制 + Token 摘要 Copy button on JSON blocks, token usage bar on message pairs JSON 块复制按钮，消息请求的 token 用量摘要栏
Session view 会话视图 Split-pane: wire view + reconstructed conversation side by side 分屏模式：原始请求 + 还原的对话并排显示
Codex support Codex 支持 Trace OpenAI Codex CLI through the same MITM front door 用同一套 MITM 前门追踪 OpenAI Codex CLI
Conversation dump 对话导出 Export reconstructed conversations as Markdown or JSON 将还原的对话导出为 Markdown 或 JSON
Token metrics Token 指标 Cumulative usage, cache hit rates, cost estimates 累计用量、缓存命中率、费用估算
Agent skill Agent Skill Claude Code skill / MCP server for programmatic trace queries Claude Code skill / MCP 服务器，编程式查询捕获的流量
Multi-session 多会话 Path-based routing: /:project/:session 基于路径的路由：/:project/:session