Claude Code Trace CLI
Reads Claude Code session logs as a conversation: tool calls, token counts and a live tail of a running session.
- 385 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The program reads every Claude Code log, which holds source code, prompts and command output.
- The efficiency analysis sends a session summary to the external Jev service, though only after confirmation.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/delexw/claude-code-trace/main/script/install-macos.sh | bashInstalls Claude Code Trace.app into `/Applications`, Apple Silicon only. For Linux and Windows use the builds from the releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The program reads every Claude Code log, which holds source code, prompts and command output.
- The efficiency analysis sends a session summary to the external Jev service, though only after confirmation.
This is third-party code. Review the repository files before installing.
What it does
Claude Code Trace turns `~/.claude/projects/*.jsonl` logs into a readable conversation. Tool calls, thinking, results, subagents, token counts and timestamps expand in place. File edits appear as word-level diffs, and MCP calls get readable names. An open session updates while Claude Code writes to its log, and the session list marks those still running. The program runs as a desktop app, a web app or a terminal UI, all reading from one Rust backend. An optional efficiency analysis sends a summary to the external Jev service only after you confirm.
Who it is for. Developers who want to see what Claude Code did in a long session and where the tokens went.
Good fit when
- The agent finished and the terminal kept no history, so you need to see which tools it called.
- A running session needs watching in a separate window while you work in another.
- Token spend in one session looks too high.
- The log contains MCP calls and you want to see which server received what.
Not a fit when
- Material for publication: the program shows sessions but does not build a file for readers.
- Work that is neither Claude Code nor Antigravity: there are no other built-in session sources.
- Analysis that must not send data out: the efficiency check sends a summary to Jev.
Example request
Install Claude Code Trace and show which tools the agent called in yesterday's session for the api project.Limitations
A one-command installer exists only for macOS on Apple Silicon: the app is ad-hoc signed and not notarized, so on first launch macOS asks you to confirm opening it in settings. Linux and Windows builds are in the releases (`.deb`, `.rpm`, `.AppImage`, `.msi`, `.exe`). Building from source needs Rust, Node.js 18 and Python 3.11 for the terminal UI. The Antigravity source is off by default and reads conversations read-only. The Jev key and provider keys live in the operating system's credential store.
How to disable. Quit the app, or stop the container with `docker compose down` if you used Docker. Remove the app from `/Applications` by moving it to the trash. Claude Code logs stay untouched.
FAQ
Which command installs Claude Code Trace on macOS?
One README command: `curl -fsSL https://raw.githubusercontent.com/delexw/claude-code-trace/main/script/install-macos.sh | bash`. It puts the app in `/Applications` and works on Apple Silicon only.
Where does session data go?
Nowhere until you run the efficiency analysis. Then a redacted summary goes to Jev, and the program shows the exact payload and waits for confirmation first.
Can it run in a browser?
Yes. `docker compose up --build` from the repository directory starts web mode at `http://localhost:1421`.
What are session sources?
They are plugins that read logs. Claude Code is built in and always on, Antigravity is enabled in Settings under General, Sources.
Is Python required?
Only for the terminal UI when building from source, where Python 3.11 or newer is needed. The desktop app and web mode do not require it.
Discussion
No questions or reviews yet.
Related
CLI toolsA debugging tool for Claude Code that reads the session logs already saved on disk and shows exact file paths, diffs, model thinking and a category-by-category
Rust terminal client claude-rs for Claude Code: the session stays in terminal scrollback, output does not flicker, tmux works.
A CLI turns coding-agent logs into an interactive HTML replay with tool calls, file diffs and navigable chapters.
A terminal explorer for Claude Code configuration: CLAUDE.md, commands, subagents and settings.json with a preview
Draws a Claude Code or Codex session as a graph of agents and tool calls, live or replayed from a recording
A Claude Code plugin that generates and maintains CLAUDE.md: a 0 to 100 quality score, a 150-line cap and stale reference checks
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Claude Code Trace (this card) | no | 385 | |
| claude-replay | no | 856 | builds a standalone HTML file for publishing, while cctrace follows a running session live |
| Claude Code History Viewer | no | 2,303 | browses history from many agents, while cctrace dissects tool calls, tokens and subagents in Claude Code sessions |
README in short
Every calling client registers with the backend and holds its own HS256 token, which can be revoked on its own. Design notes for each part live in the repository's specs folder.