claude-devtools

The second major feature group encompasses the Tool Call Inspector and enhanced copy-paste functionality. The Tool Call Inspector expands every collapsed action, such as 'Read 3 files', into a syntax-highlighted, detailed view showing the exact content read, inline diffs for Edit calls, and the full output from Bash or Grep commands. Alongside this, the application solves the terminal's awkward text handling by rendering every message, tool call, and output as real, selectable text without ANSI code leaks or selection wrapping issues, allowing one-click copying of code blocks, file paths, or entire responses. This combination ensures developers can not only see the precise actions but also efficiently extract and reuse the generated content across macOS, Linux, and Windows platforms.

Additional capabilities include Project Memory access, Team & Subagent Trees, and Notification Triggers. The Project Memory pane opens the MEMORY.md index and associated layer files (like working style or architecture notes) stored in `~/.claude/projects/`, providing full markdown rendering with in-pane search and a launcher to open files in external editors. The Team & Subagent Trees feature isolates execution traces per agent, showing nested, recursive trees of tool calls with individual token metrics, duration, and cost calculations. Notification Triggers offer system alerts for specific events like .env file access, tool execution errors, or high token usage, and can be customized with regex patterns to monitor sensitive file paths or set thresholds on various token counts, adding a layer of security and monitoring.

The product works by directly reading and parsing the structured session log files that Claude Code automatically writes to the `~/.claude/` directory on the user's local disk after every session. It does not act as a wrapper or intermediary for Claude Code itself, does not make any outbound API calls, and requires no login or API keys. The application reconstructs the session by decoding the raw JSON data that the terminal normally hides or only shows as verbose noise, organizing it into a coherent, visual interface with features like a command palette for cross-session search, a multi-pane layout for comparing sessions, and even SSH support for inspecting logs on remote machines by reading `~/.ssh/config` and supporting agent forwarding.

Key benefits and measurable outcomes for users include restored visibility and control over the AI coding process, leading to better debugging, learning, and security auditing. Developers can precisely see where tokens are being consumed, understand costly or inefficient tool calls, and verify that the AI is not accessing sensitive files like payment or billing directories. The structured, searchable interface saves significant time otherwise spent sifting through verbose JSON dumps, and the copy-paste functionality eliminates friction when reusing generated code. Ultimately, it transforms Claude Code from a black-box code generator into a transparent, analyzable collaborator whose entire workflow can be reviewed and optimized.

Concrete use cases involve specific workflow examples such as a developer auditing a session where Claude unexpectedly modified a critical configuration file; using the Tool Call Inspector and context breakdown, they can trace back to the exact thinking step and file read that led to the change. Another example is a team lead comparing sessions side-by-side in the multi-pane layout to understand why two similar prompts produced different results, analyzing the subagent coordination and token allocation differences. A security-conscious engineer can set regex notification triggers for any access to `*.env` or `*secret*` paths and receive immediate system alerts, then use the session replay to see the full context and tool output surrounding that access.

The target users are primarily software developers, engineers, and technical teams who actively use Claude Code for programming assistance and have been impacted by the loss of detailed output. The tool integrates with the existing local file system and supports workflows across all major operating systems via native installers for macOS (Apple Silicon and Intel), Windows, and Linux. Its tech stack is open-source under the MIT license, and it offers direct downloads or installation via Homebrew (`brew install --cask claude-devtools`). There is no mentioned pricing model, as it appears to be a free, open-source tool with 'No login · no API keys' and 'Zero outbound calls,' focusing on accessibility and community adoption, already used by over 50,000 developers.

In summary, claude-devtools is an indispensable forensic tool that gives developers back the visibility Claude Code took away, turning local session logs into a rich, interactive audit trail. By reconstructing every thought, tool call, and token from the data already on disk, it empowers users to understand, trust, and effectively collaborate with AI coding assistants, ensuring they are no longer coding blind but can see and control every aspect of the AI's work in their development environment.

Updated 2026-02-28