When you run Claude Code in several terminals at once, it quickly becomes hard to tell which session is doing what and how far it has gotten. Claude Task Viewer is an open-source dashboard that pulls that picture together onto a single board. It watches the directory where Claude Code writes its task data (~/.claude/tasks/) and renders states such as in-progress and completed on a Kanban-style board in the browser. Running npx claude-task-viewer starts a local server, and opening http://localhost:3456 is all it takes to begin. It is built by L1AD, licensed under MIT, and was released in January 2026.
Its design principle is “Observation over Control.” You cannot rewrite task states or descriptions from the board — the only human actions are deleting a task and adding a note. It helps to think of it as a mirror that reflects Claude Code’s actual state rather than a tool that drives it.
Key Features
- Real-time updates with no polling: Task state changes are pushed to the browser over Server-Sent Events. You can watch work progress without reloading the page
- Cross-session view: Sessions under
~/.claude/tasks/are detected automatically and laid out on one board, so you can see the whole picture without cycling through terminal tabs - Dependency visualization:
blockedBy/blocksrelationships are displayed so you can see which task is waiting on what. Deletion includes a dependency check - Completion notifications (desktop alerts and sound): A desktop notification and a chime fire when a task finishes, so you can wait for long-running work while doing something else
- Gantt-style timeline: A timeline view with an auto-scaling time axis shows when each task in a session started and how long it took
- Automatic archiving of stale sessions: Sessions with no changes for more than seven days and no active tasks are collapsed into an archive area. It is hidden by default but still searchable
- Fuzzy search and keyboard shortcuts: Fuzzy search spans session names and task descriptions.
?opens help,Ddeletes the selected task, andEsccloses panels
Startup options are available as well: PORT=8080 changes the port, --open launches the browser automatically, and --dir=~/.claude-work points the viewer at a different Claude configuration directory.
Pricing
| Plan | Price | Highlights |
|---|---|---|
| Open source (MIT) | $0 | All features available. Runs locally with no account required |
Pricing information is current as of August 2026. Please check the official repository for the latest details.
There is no paid plan or hosted version. You need Node.js locally, plus Claude Code itself to generate the tasks being observed.
Pros & Cons
✅ Pros
- Starts with a single
npxcommand, with almost no installation or configuration files to prepare - Everything runs locally, so task information is never sent to an external service
- Makes it immediately clear which of your parallel sessions is currently active
- Completion notifications mean you no longer have to sit watching the terminal for long jobs to end
- MIT-licensed open source, so you can read the implementation whenever behavior is unclear
⚠️ Cons
- Observation only — you cannot create tasks or move their states from the board
- Assumes Claude Code is writing the task files; other coding agents are not supported
- Built on local file watching, so it is not suited to aggregating sessions running on remote machines
- An individually developed tool without permission management aimed at large team use
Comparison with Similar Services
| Criteria | Claude Task Viewer | claude-code-kanban | Agetor | Terminal only (default) |
|---|---|---|---|---|
| Main role | Observing and visualizing tasks | Visualizing tasks and agents | Dispatching and orchestrating tasks | Checking output per session |
| Scope of actions | Delete and notes only | Mostly viewing | Trigger runs from the board | Direct command operation |
| How to start | npx claude-task-viewer | npx claude-code-kanban | Self-hosted | No extra tool needed |
| Update method | Server-Sent Events | Server-Sent Events | ─ | ─ |
| License | MIT (open source) | Open source | Open source | ─ |
All of these start from Claude Code’s local files. The difference lies in whether a tool sticks to observation or also drives execution from the board. Claude Task Viewer commits fully to the former, which is exactly why it is so light to adopt.
Who Is It For
- Developers who run two or three or more Claude Code sessions at the same time
- People who kick off long-running tasks and want to work on something else while waiting
- Anyone who wants an overview of where a chain of dependent tasks is stuck
- Those who want local-only monitoring without handing logs to an external service
- Anyone who wants to try visibility first, without paying an extra learning cost
Summary
Claude Task Viewer is a deliberately narrow tool: it reads Claude Code’s task data and shows it on a Kanban board. In exchange for giving up any ability to change state, it gains the convenience of a single npx launch and a clear view across sessions. If parallel work has grown to the point where terminal tab management is breaking down, it is worth a try. If you instead want to dispatch tasks and control execution from the board, an orchestration-style tool is the better fit.