Claude Devtools
The missing DevTools for Claude Code — inspect session logs, tool calls, token usage, subagents, and context window in a visual UI. Free, open source.
Last verified:
What is Claude Devtools?
claude-devtools is a free, open-source debugger for Claude Code that reads session logs from ~/.claude/ on your machine and reconstructs full session transcripts. It reveals everything Claude Code's terminal hides: every file path Read, every tool call with exact inputs and outputs, thinking steps, token usage breakdowns, subagent execution trees, and context window compaction details. The tool does not wrap, modify, or interact with Claude Code itself—it is purely read-only observability layer working with logs that already exist.
Key features include context reconstruction with per-turn token attribution across 7 categories (CLAUDE.md files, skill activations, @-mentioned files, tool I/O, thinking, team overhead, user text), a tool call inspector with syntax-highlighted Read calls and inline Edit diffs, project memory viewing with dedicated panes, team and subagent trees with nested execution visualization, notification triggers for .env access and errors, SSH remote session support, command palette with cross-session search, multi-pane layout for parallel session comparison, and clean copy-paste without terminal ANSI codes or line-wrap issues.
The tool is designed for developers using Claude Code CLI who need visibility into what their AI assistant actually did. It is used by 50,000+ developers, requires no API keys or login, works on macOS (Apple Silicon and Intel), Windows, Linux, and Docker, and supports every session you've ever run from any tool. It is the structured, navigable alternative to the --verbose flag which dumps raw JSON noise.
Claude Devtools pricing
Pricing model: Freemium
Free. claude-devtools is completely free and open source under the MIT license. No paid plans, no tiers, no APIkey costs. The tool requires no login and makes zero outbound calls.
Claude Devtools pros
- Free and open source with MIT license
- No API keys required
- No login required
- Zero outbound calls—100% local
- Read-only—does not modify Claude Code
- Works with every session already on disk
- Per-turn token attribution across 7 categories
- Syntax-highlighted Read calls
- Inline Edit diffs visualization
- Full subagent execution trees with nesting
- Context window compaction visualization
- Built-in notifications for .env access and errors
- SSH remote session support with SFTP
- Cross-session search via command palette
- Multi-pane layout for parallel session comparison
- Clean copy-paste without terminal ANSI codes
- Project memory pane with in-pane search
- Supports macOS, Windows, Linux, and Docker
- Homebrew installation on macOS
- Opens files in Cursor, VS Code, Zed, Finder
Claude Devtools cons
- macOS-focused installation (Homebrew cask)
- Requires Claude Code session logs to already exist
- Only works with Claude Code, not other AI tools
- Post-hoc viewer—not live during session
- No cloud sync or collaboration features
- Desktop app only—no web interface
- SSH support requires manual config setup
- Limited to local file system access
Frequently asked questions about Claude Devtools
What is claude-devtools?
claude-devtools is a free, open-source Claude Code debugger that reads ~/.claude/ logs and reconstructs full session transcripts—every file path, tool call, thinking step, and token—in a structured, searchable interface. It does not wrap or modify Claude Code.
Where does Claude Code save session logs?
Claude Code saves session transcripts as JSONL logs to ~/.claude/ on your machine. On macOS: /Users/<you>/.claude/, on Linux: /home/<you>/.claude/, on Windows: C:\Users\<you>\.claude\. These logs exist whether or not you use claude-devtools.
How do I view a session?
Open claude-devtools and select the session from the sidebar. You'll see the full session transcript with every tool call expanded, thinking steps visible, token usage broken down per turn, and subagent execution rendered as a tree.
What format are Claude Code transcripts in?
Claude Code transcripts are stored as JSONL files in ~/.claude/projects/. Raw transcripts are unreadable with escaped JSON and system prompts. claude-devtools parses them and renders a chronological conversation view with specialized renderers for each tool and cross-session search.
Does claude-devtools interact with Claude Code?
No. claude-devtools is read-only. It reads the session logs that already exist on your machine and has zero interaction with Claude Code itself. It works with every session you've ever run.
What platforms are supported?
macOS (Apple Silicon and Intel), Windows, Linux, and Docker. Install via Homebrew (brew install --cask claude-devtools) or download directly from GitHub releases.
How does token attribution work?
claude-devtools provides per-turn token attribution across 7 categories: CLAUDE.md files, skill activations, @-mentioned files, tool call I/O, extended thinking, team overhead, and user text. This replaces the terminal's three-segment progress bar with actual numbers.
What is the difference between --verbose and claude-devtools?
Both surface data Claude Code's terminal hides. --verbose prints the raw event stream as JSON to stdout during the run. claude-devtools reads the JSONL transcripts after they're written and renders them as a structured conversation with per-tool viewers, navigable history, and search.
Can I inspect remote sessions?
Yes. claude-devtools supports inspecting Claude Code sessions on remote machines via SSH. It reads ~/.ssh/config, supports agent forwarding and key auth, and lets you switch between local and remote instantly.
How do I access project memory?
Per-project Claude memory lives at ~/.claude/projects/<project>/memory/—MEMORY.md as an index plus one .md per layer. claude-devtools opens this in a dedicated pane with layer list on the left, full markdown rendering with in-pane search on the right, and an Open in launcher for Finder, Cursor, VS Code, and more.