claude-kanban 0.1.0

A real-time kanban dashboard for monitoring Claude Code sessions across multiple servers

Claude Code Agent Kanban

A real-time web dashboard that monitors Claude Code sessions across multiple servers, displaying them as a kanban board with AI-powered task summaries.

If you run Claude Code on multiple machines simultaneously, it's easy to lose track of what each agent is doing. This tool SSHs into your servers, collects all active sessions, and presents them in a single unified view.

Features

• Multi-server monitoring — SSH into remote servers to collect session data in parallel
• Live kanban board — Sessions organized into Running / Completed / Errors columns
• AI-powered summaries — Each session is summarized by Claude (via local claude CLI) with task description, progress status, and estimated completion percentage
• Auto-refresh — Dashboard updates automatically; fast-polls on first load to pick up AI summaries quickly
• Session lifecycle tracking — Sessions move from Running to Completed when the Claude Code process exits; resumed sessions move back to Running
• Web-based configuration — Add/edit/remove SSH servers and test connections from the Settings panel, no config files needed
• Summary caching — Summaries are cached on disk; unchanged sessions won't be re-summarized, even across restarts

Prerequisites

• Python 3.11+
• uv package manager
• Claude Code CLI installed (used for AI summaries via claude -p)
• SSH key-based access to remote servers (for multi-server monitoring)

Quick Start

Install with uv (recommended)

uvx claude-kanban

Run from source

git clone https://github.com/madsys-dev/claude-kanban.git
cd claude-kanban
uv run claude-kanban

Open http://localhost:5555 in your browser.

By default, it scans the local machine for Claude Code sessions. To add remote servers, click Settings in the top-right corner.

How It Works

1. Session discovery — Scans ~/.claude/sessions/*.json on each server to find active Claude Code processes
2. Conversation parsing — Reads the corresponding JSONL conversation logs from ~/.claude/projects/ to extract message history
3. AI summarization — Sends conversation excerpts (first 3 + last 6 messages) to claude -p --model haiku for summarization
4. Lifecycle tracking — When a session's PID disappears, it moves to the Completed column. If resumed, it moves back to Running

Configuration

All configuration is done through the web UI (Settings button). Under the hood, settings are persisted to ~/.claude-kanban/config.yaml:

include_local: true
servers:
  - host: gpu-server-1.example.com
    user: ubuntu
    label: GPU Server 1
  - host: 10.0.0.50
    user: root
    port: 2222
    key: ~/.ssh/id_ed25519
    label: Dev Box

Server options

Field	Required	Default	Description
host	Yes	—	Hostname or IP
user	No	Current user	SSH username
port	No	22	SSH port
key	No	~/.ssh/id_rsa	Path to SSH private key
label	No	Same as host	Display name on the dashboard

Remote server requirements

• Python 3 installed
• Claude Code installed (sessions stored in ~/.claude/)
• SSH key-based authentication configured

Environment variables

Variable	Default	Description
KANBAN_CONFIG	~/.claude-kanban/config.yaml	Path to configuration file
KANBAN_DATA_DIR	~/.claude-kanban	Data directory for config and summary cache

API

Endpoint	Method	Description
GET /api/sessions	GET	Returns all sessions (cached 30s)
POST /api/refresh	POST	Force refresh and return sessions
GET /api/servers	GET	List configured servers
POST /api/servers	POST	Add a server
PUT /api/servers/<id>	PUT	Update a server
DELETE /api/servers/<id>	DELETE	Remove a server
POST /api/servers/<id>/test	POST	Test SSH connection
PUT /api/config/local	PUT	Toggle local machine scanning

License

MIT