DockTerm
Run Claude Code in a desktop terminal workspace and keep its progress, changes, and permission requests in view.
- Source repo
- munvard/dockterm
- Stars
- ★ 29
- Last updated
- 1d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 65/100 · Some gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Developers who want several Claude Code project terminals in one window, with each pane tied to a different repository.
- Not a fit if
- Teams that need a coding assistant other than Claude Code
- Users who cannot accept an unsigned Windows installer
- Source review
- 65/100 · Some gaps
What does this agent do, and when should you use it?
DockTerm is an Electron desktop app for macOS, Windows, and Linux that gives Claude Code a terminal-first workspace. It runs your real `claude` session in a shell backed by a native PTY, with on-demand panels for files, editing, Git, diffs, checkpoints, and MCP. Its on-screen mascot, munu, reflects Claude's terminal state and can surface permission prompts for you to answer. DockTerm reads usage and activity information from local Claude files and transcripts; it says it uploads no data, runs no AI of its own, and never answers permission prompts automatically. It suits developers who want to keep using Claude Code while managing projects and reviewing changes in one window; it does not suit teams looking for a different coding assistant.
DockTerm starts a real shell and PTY terminal so you can run claude or claude --resume in a project. It displays Claude session checkpoints, subagent activity, and selected terminal text, which can be sent to Claude as a referenced snippet; you can review diffs since the last commit, the current session, or a pinned checkpoint, then stage, commit, push, or pull Git changes. A file tree, Quick Open, file-content search, Monaco editor, and image previews support project browsing and editing. Read-only panels show MCP servers, skills, slash commands, and subagent information, with secrets masked. munu reflects Claude's state from terminal output and surfaces permission prompts for you to answer; it does not answer them for you. DockTerm also reads Claude-reported five-hour and weekly usage percentages and reset times from local ~/.claude files.
- Developers who want several Claude Code project terminals in one window, with each pane tied to a different repository.
- Developers who need to review Claude's diffs and stage or commit changes through a confirmed Git interface.
- Users who want to see Claude's permission prompts while away from DockTerm or using a fullscreen app.
- Developers who need to monitor Claude Code subagents, their elapsed time, and result previews.
- Claude Code users who want local views of usage percentages, reset times, and an even-pace marker.
How do you install or deploy this agent?
On macOS, install with Homebrew:
brew install --cask munvard/dockterm/docktermYou can also download a platform build from GitHub Releases: a DMG for Apple Silicon or Intel Macs, an EXE for Windows 10/11, or an x86-64 AppImage for Linux. The README says macOS builds are signed and notarized; Windows builds are currently unsigned and may show a SmartScreen warning. To build from source, you need Node.js 22 or later:
git clone https://github.com/munvard/dockterm && cd dockterm
npm install
npm run devHow do you use this agent?
Install and open DockTerm, choose or open a project, then run Claude Code in its terminal pane:
claudeTo continue an existing session, run:
claude --resumeUse the panels to inspect files, Git status, diffs, checkpoints, and MCP configuration. When munu surfaces a permission prompt, you choose an answer. Showing local Claude usage requires allowing DockTerm to read ~/.claude; usage and session data are described as read-only.
What are this agent's strengths and limitations?
- Keeps the real
claudeshell session, so there is no move to a separate chat interface. - Adds diff review and Git stage, commit, push, and pull actions with confirmation for risky operations.
- munu can surface permission prompts over fullscreen apps or on another desktop, and never answers them automatically.
- Reads Claude usage, checkpoints, and agent activity locally; the project states it has no telemetry, accounts, or AI of its own.
- Its core workflow is built around Claude Code, so it does not fit teams that need another coding assistant.
- The Electron app has a fairly large download; the README describes the project as early and actively developed, with bugs and rough edges expected.
- The Windows build is currently unsigned and may trigger a SmartScreen warning.
- Building from source requires Node.js 22 or later; MCP, checkpoint, and usage views rely on local configuration or Claude files.
How does this agent compare with similar options?
The README compares DockTerm with a raw terminal, a full IDE, and Claude Code GUIs: DockTerm keeps the real Claude Code session while adding diff review, Git actions, and munu status prompts. It names Claudia and Opcode as examples of Claude GUIs that replace the terminal.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| DockTerm This agent | 65 · Some gaps | Desktop appFree + model costs | ★ 29 | 1d ago | TypeScript | Claude Code |
| Mirafold | 89 · Good | CLIFreemium | ★ 18 | 26d ago | TypeScript | Codex · Claude Code |
| Calyx | 59 · Major gaps | Desktop appFree | ★ 346 | 2d ago | Swift | Codex · Claude Code |
| Kun Workspace | 49 · Major gaps | Desktop appFree + model costs | ★ 6.3k | 1d ago | TypeScript | ChatGPT · Codex · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
README and SECURITY.md describe local operation, no telemetry, read-only MCP configuration with masked secrets, project-scoped filesystem access, IPC validation, disabled Git hooks, and confirmation for risky Git actions, giving ordinary permissions and data flows stated boundaries. Deductions: these are mostly documentation claims; the supplied material lacks implementation details to verify them. The dependency list includes no vulnerability review evidence, and rollback guidance is limited. The repository names an author and MIT license, but publisher identity is explicitly unverified and cannot be confirmed from that metadata.
README, package.json, and CI align on Node 22, typechecking, unit tests, and builds across three platforms; supplied tests show parser and activity-reducer cases. Deductions: this static review cannot establish execution results, and the files show little about user-facing failure messages or recovery behavior.
README targets Claude Code users, describes terminal, review, Git, file, and MCP workflows, states that it wraps a real Claude session and does not auto-answer permission prompts, and lists macOS, Windows, and Linux. Deductions: the supplied material says little about use without Claude Code or more unusual environments, and platform support is primarily asserted in the README.
README organizes features, shortcuts, installation, privacy and security, FAQ, source builds, and status, with screenshots, animations, and an MIT license. It discloses installation steps and limitations including unsigned Windows builds, early-stage status, and Electron size. Deductions: the supplied files include no changelog or clear release cadence; maintenance responsibility is only partly indicated by author metadata, issue/PR invitation, and a claim of active development.
The materials describe a workspace combining a real terminal, Claude state prompts, diff review, and project tools, with a readable feature list, shortcuts, and screenshots. Deductions: value and convenience are primarily README claims; the static materials do not establish realized gains, resource costs, or net benefit across users.
Several privacy and security claims in README are echoed in SECURITY.md; package.json and CI list build, typecheck, and test commands, and supplied tests contain concrete behavior cases. Deductions: implementation files are absent, so most product and security claims cannot be checked directly; a test-fixture comment says its schema was checked against real transcripts, but the supplied material does not corroborate that independently.
- Security boundaries, no-telemetry behavior, and local data handling are primarily documented claims; the supplied files do not expose enough implementation to inspect them.
- The README says Windows builds are currently unsigned and may trigger an unknown-publisher SmartScreen prompt.
- The project identifies itself as pre-1.0 with expected rough edges; security fixes target the latest main release and are not backported.
FAQ
Does DockTerm itself cost money?
Does DockTerm upload code or session data?
~/.claude files as read-only data; MCP and skills configuration is also presented read-only.