matt1398/claude-devtools

What it solves

Claude Code often hides critical details in the terminal, such as exact file paths, chain-of-thought reasoning, and detailed tool call inputs/outputs. This makes debugging and auditing AI-generated changes difficult. claude-devtools provides a structured, navigable interface to see everything the AI did without relying on raw JSON dumps from --verbose mode.

How it works

Instead of acting as a wrapper or modifying the AI tool, it reads the session logs and transcripts already stored locally on your machine (typically in ~/.claude/). It reconstructs these logs into a visual interface where you can inspect tool calls, token usage, and agent activity.

Who it’s for

Developers using Claude Code who need to audit the AI's reasoning process, track token costs, or recover specific code blocks and file changes that were collapsed in the terminal output.

Highlights

• Tool Call Inspector: View syntax-highlighted file reads, inline diffs for edits, and full bash output.
• Context Reconstruction: Per-turn token attribution across categories like thinking, tool I/O, and project memory.
• Project Memory Viewer: A dedicated pane to browse and open MEMORY.md and its associated layers in your preferred editor.
• Subagent Trees: Recursive visualization of agents spawning other agents, including their specific tool traces and costs.
• Compaction Visualization: Visual tracking of how the context window fills and compresses over time.
• SSH Remote Support: Ability to inspect sessions hosted on remote machines via SSH.