# kagan-sh/kagan [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/kagan-sh/kagan  
**GitHub Stars:** 10  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/kagan-sh-kagan

## Description
AI-powered Kanban TUI and MCP server for autonomous development workflows. Orchestrates 14 coding agents across task tracking, isolated git worktrees, review, and merge.

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "kagan": {
    "command": "npx",
    "args": ["-y","kagan-sh-kagan"]
  }
}
```

## Documentation

## What kagan-sh/kagan MCP server does

The kagan-sh/kagan MCP server organizes AI coding work as supervised tasks inside OpenCode. It presents work on a kanban board with four stages: Backlog, In Progress, Review, and Done. Each transition has a gate, so a task is not treated as complete merely because an agent has produced code.

The workflow is designed for projects where multiple coding tasks need tracking, isolation, review, and an explicit merge decision. Kagan keeps agent changes out of the primary checkout while the work is in progress, then requires review and triage before the task can reach Done.

## How it works

Every task becomes an OpenCode session associated with an isolated git worktree. The agent works on a `kagan/<slug>` branch rather than directly in the main checkout. This gives each task its own working area and branch while preserving the repository’s regular checkout for other work.

The process begins with intake before the coding agent runs. After implementation, a reviewer agent evaluates the result against the original task and records ranked findings. Before completion, the user must review those findings, decide how to handle each one, and choose whether and where the changes should be merged. Kagan therefore supports an approval-oriented workflow rather than unattended promotion from implementation to completion.

## Setup and configuration

Kagan runs as an OpenCode plugin and requires OpenCode 1.17.13 or newer. The documented global installation command is:

```bash
opencode plugin -g @kagan-sh/kagan
```

A local clone can instead be added to both OpenCode configuration files through the `plugin` array. Once installed, open the board with `/kagan`, the `kagan` command-palette entry, or the `<leader>k` shortcut. The default leader key is `ctrl+x`.

Tasks can also be created conversationally from a regular OpenCode session with `/kagan-task`. Project settings are available through `/kagan-settings`, and the README directs users to the configuration reference for option syntax.

For global npm installations, Kagan checks npm for newer stable releases and displays available updates in the board footer. Updates can be reviewed with `u`, `/kagan-update`, or the command palette. Local, file, and development installations are not updated automatically.

## Tools and capabilities

The kagan-sh/kagan MCP server supports these documented workflow capabilities:

- Create and manage supervised coding tasks on a kanban board.
- Run each task in an isolated git worktree and `kagan/<slug>` branch.
- Start tasks only after intake review.
- Submit completed work to a reviewer agent.
- Record ranked review findings against the original task.
- Require finding triage and an explicit merge decision before Done.

The supplied material does not enumerate individual MCP tool names or describe external service integrations. Its documented integration point is OpenCode.

## Limitations and notes

Kagan is tied to OpenCode and requires version 1.17.13 or newer. The README describes it as an OpenCode plugin as well as an MCP server, so it should be evaluated primarily for OpenCode-based development workflows. Agents do not work directly in the main checkout; task changes are made in isolated worktrees and branches instead.

Review completion is intentionally not automatic. A reviewer can file findings, but the user must triage every finding and decide whether or where to merge. The material does not specify support for other coding environments, hosted deployment, authentication settings, or named MCP clients.

The project is released under the MIT license.

_Full upstream README: https://allmcps.com/mcp/kagan-sh-kagan/readme_