claudedash was an experimental dashboard for Claude Code, featuring real-time Kanban, MCP server for self-querying, git snapshot/recover, quality gates, and more. However, I've decided to integrate its best features into a more mature and actively maintained project: siteboon/claudecodeui.
Why the switch?
- Better community & features: siteboon/claudecodeui has 7k+ stars, desktop/mobile support, and already includes similar tools like TaskMaster Kanban and Git Explorer.
- Benefits for you: Switch to siteboon for a more robust experience with ongoing updates.
If you're using claudedash, migrate now! Install siteboon/claudecodeui via npm i claudecodeui or check their README for setup.
Thanks for the support – let's build better AI tools together! 🚀
Local control plane for AI coding agents: recover, guard, and steer execution.
You tell Claude Code: "refactor the auth system." It says: "on it."
Terminal scrolls. Minutes pass. Is it on step 3 of 12? Stuck in a loop? Already done? Context almost full?
You have no idea.
claudedash fixes that with a local control plane:
- State + safety first (queue state, context risk, quality gates)
- Recovery first (snapshot + recover)
- UI optional (API-first, browser view when needed)
npx -y claudedash@latest startNeed the web UI immediately?
npx -y claudedash@latest start --openThis is for you if:
- You run Claude Code agents on long tasks (>5 minutes) and need visibility
- You've lost work because context filled up and the agent went silent
- You want structured task tracking with dependencies and acceptance criteria
This is NOT for you if:
- You only do short, single-turn Claude Code interactions
- You need multi-user team infrastructure (this is a local-first developer tool)
# Zero-install — always gets the latest version
npx -y claudedash@latest start
# Open the companion web UI (optional)
npx -y claudedash@latest start --open
# Install lifecycle hooks (recommended)
claudedash hooks install
# Set up plan mode for structured task execution
claudedash initThat's it. The control plane auto-detects your Claude session files.
- Plan mode — structured execution with dependencies, acceptance criteria, and blocked-task detection
- Context rollback — snapshot + recover to restore code and execution context together
- Quality gates — lint / typecheck / test results per task with full timeline history
- Context health — token risk warnings at 65% (warn) / 75% (critical)
- Agent API —
POST /log,POST /agent/register, heartbeat, task automation endpoints - Hook integration — PostToolUse/Stop/PreCompact/PostCompact hooks stream live events
- Live session monitor — optional web UI over
~/.claude/tasks/+~/.claude/todos/ - MCP server —
get_queue,get_sessions,get_cost,log_taskfor agent self-query - Zero infra — no database, no cloud, reads files from
~/.claude/directly
| Live Mode | Plan Mode | |
|---|---|---|
| What | Watch Claude work in real time | Structured project execution |
| Source | ~/.claude/tasks/ |
.claudedash/queue.md |
| Setup | None | claudedash init |
| Use when | Visibility matters | Control matters |
| Deps / AC | — | ✓ Full dependency graph |
| Execution log | — | ✓ execution.log |
Live mode is on by default. Both modes can run simultaneously.
 Live View — Session Kanban + context health per agent |
 Queue — Plan mode task board with dependency graph |
 Worktrees — Parallel agents across git branches |
 Activity — Cost tracker + tool analytics per session |
 Context Health — Live token usage, warnings at 65% / 75% |
 Quick Start — One command, zero config |
# Zero-install, always latest
npx -y claudedash@latest start
# Or install globally
npm i -g claudedash
claudedash startInitialize in your project:
claudedash initThis creates .claudedash/ with:
| File | Purpose |
|---|---|
queue.md |
Task list — slices, dependencies, acceptance criteria |
workflow.md |
Execution protocol for your agent |
execution.log |
Agent logs DONE / FAILED / BLOCKED here |
CLAUDE.md |
Paste into your project's CLAUDE.md |
queue.md format:
# Slice S1
## S1-T1
Area: Backend
Depends: -
Description: Setup database schema
AC: Tables created, migrations run
## S1-T2
Area: Backend
Depends: S1-T1
Description: Implement user auth
AC: Login and registration workingTell your agent:
Follow .claudedash/workflow.md, start with S1-T1.
| Command | Description |
|---|---|
claudedash start |
Start control plane API (auto-detect modes) |
claudedash start --open |
Start API and open companion web UI |
claudedash start -p 3000 |
Custom port |
claudedash start --source claude-code |
Select source adapter (current: claude-code) |
claudedash start --token <secret> |
Enable auth token |
claudedash init |
Init plan mode in current directory |
claudedash hooks install |
Install PostToolUse/Stop/PreCompact hooks |
claudedash hooks install --git |
Also install git post-commit auto-snapshot |
claudedash status |
Single-line terminal summary (no browser) |
claudedash doctor |
Check setup: hooks, port, version, queue |
claudedash snapshot --commit |
Save context snapshot tagged to HEAD commit |
claudedash snapshots list |
List all saved snapshots |
claudedash recover |
Summarize last session after /clear |
claudedash recover <hash> |
Recover context from a specific commit |
By default claudedash listens on 127.0.0.1 only.
# Start with a token
claudedash start --token mysecret123
# Or via env
CLAUDEDASH_TOKEN=mysecret123 claudedash startTeam access — use Authorization: Bearer header:
# In browser via tunnel (recommended)
ngrok http 4317
# → Share the ngrok URL with trusted teammates; API access requires Bearer token
# Via curl / API clients
curl -H "Authorization: Bearer mysecret123" http://your-host:4317/sessionsWhen you open the web UI with token auth enabled, claudedash shows a token sign-in screen and stores a short-lived auth cookie for browser requests.
Tip: Use a tunnel instead of exposing
--host 0.0.0.0:claudedash start --token $(openssl rand -hex 16) ngrok http 4317Security:
--host 0.0.0.0requires--token. Token is validated viaAuthorization: Bearerheader only — never via query string.
Color-coded token usage per session — green → yellow → red at 65% / 75%. → Context Health docs
Log lint, typecheck, test results per task via meta.quality in execution.log.
See ✅/❌ inline in each card, with full timeline.
→ Quality Gates docs
Running agents across multiple git branches? The Worktrees tab maps sessions to branches by cwd, shows dirty/ahead/behind state, and lists which tasks are running where. Native support for claude --worktree <name> (creates .claude/worktrees/<name>/).
→ Worktree docs
Something broke during vibe coding? Roll back your code and your Claude context together.
# 1. Auto-snapshot on every commit (one-time setup)
claudedash hooks install --git
# 2. When something breaks, find the last good commit
claudedash snapshots list
# [abc1234f] Feb 23 14:02 main 12/15 done
# 3. Roll back code + context in two commands
git reset --hard abc1234f
claudedash recover abc1234fThe Snapshots tab in the dashboard shows all checkpoints with a one-click "Copy rollback" button.
Claude can query its own control plane:
claude mcp add claudedash -- npx -y claudedash@latest mcp
# If your claudedash server uses --token:
claude mcp add claudedash -- npx -y claudedash@latest mcp --token "$CLAUDEDASH_TOKEN"Tools: get_queue, get_sessions, get_cost, get_history, log_task, create_task, register_agent, send_heartbeat.
| Endpoint | Description |
|---|---|
GET /health |
Status, modes, connected clients |
POST /auth/login |
Exchange token for browser auth cookie |
POST /auth/logout |
Clear browser auth cookie |
GET /sessions |
All sessions with context health |
GET /sessions/:id/context |
Session JSONL summary |
GET /events |
SSE stream |
GET /snapshot |
Plan mode state |
GET /queue |
Computed task statuses |
GET /worktrees |
Git worktrees with task associations |
GET /snapshots |
List all context snapshots |
GET /snapshots/:hash |
Get snapshot by commit hash |
DELETE /snapshots/:hash |
Delete a snapshot |
GET /billing-block |
Current 5h billing window |
GET /cost |
Estimated cost by model |
POST /log |
Log task result |
POST /plan/task |
Add task to queue.md |
POST /agent/register |
Register an agent |
claudedash reads ~/.claude/ files directly — no server integration required.
→ docs/compatibility.md — exact file paths, JSONL schema, and what breaks if Claude Code changes formats.
TypeScript · Fastify · chokidar · SSE · Next.js 16 · Tailwind CSS · Vitest
git clone https://github.com/yunusemrgrl/claudedash.git
cd claudedash && npm install
cd dashboard && npm install && cd ..
npm run build # Build core + dashboard
npm test # Run tests
npm run dev # Dev server with watchPRs welcome. Open an issue first for anything beyond small fixes. See Releases for version history.
MIT — not affiliated with Anthropic.