Connects Claude to a Kanban workflow system built around AI agent orchestration with mandatory human review gates. The MCP server exposes operations to create tasks, move cards through BACKLOG, IN_PROGRESS, REVIEW, and DONE states, and manage isolated git worktrees where agents execute work. The state machine enforces that no task reaches DONE without explicit human approval at the REVIEW stage. You'd reach for this when you want AI agents handling implementation while keeping yourself in the loop as a structural checkpoint before any code hits main. Works with Claude Code, Codex, Gemini CLI, and other MCP clients. The server sits alongside a TUI, optional web dashboard, and VS Code extension that all share the same board state.
Supervised kanban for AI coding agents inside OpenCode.
Kagan is an OpenCode plugin that turns agent work into supervised tasks on a kanban board. Each task is an OpenCode session running in an isolated git worktree, moving through Backlog → In Progress → Review → Done with a gate at every transition — intake before the agent runs, review before you approve, merge only on your say-so.
The agent never touches your checkout. It works on a kagan/<slug> branch in its own worktree, a reviewer agent files ranked findings against the original task, and nothing reaches Done until you've triaged every finding and chosen where — or whether — to merge.
You need OpenCode 1.17.13 or newer (see engines.opencode in
package.json).
Install globally so Kagan is available in every project:
opencode plugin -g @kagan-sh/kagan
Or add a local clone to both OpenCode config files:
{
"plugin": ["/path/to/kagan"]
}
Open the board with /kagan from the command palette, the kagan palette command, or <leader>k (the leader key defaults to ctrl+x).
From any regular OpenCode session, run /kagan-task to create board tasks conversationally — useful when planning several tickets at once without opening the board create dialog for each one.
For global npm installs, Kagan checks npm for newer stable releases and shows an available version in
the board footer. Press u, run /kagan-update, or use the command palette to review and confirm the
update; Kagan then installs the exact release and asks you to restart OpenCode. Local, file, and
development installs are never updated automatically. See
Updating.
Pass options by using the array-of-array form, or open /kagan-settings from the project — see the configuration reference.
Documentation is available in docs/.
KAGAN_OPENAI_API_KEYsecretOpenAI API key for agent execution (if using OpenAI backend)
KAGAN_ANTHROPIC_API_KEYsecretAnthropic API key for agent execution (if using Claude backend)
KAGAN_DB_PATHCustom path for Kagan database file
io.github.ericm1018/skillfm-llm-cost-optimizer-openai-anthropic-usage
io.github.mikerawsonnz/llm-orchestration-agent
io.github.mikerawsonnz/authenticated-llm-agent
labforgedev/copilot-memory-mcp
csoai-org/agent-prompt-injection-firewall-mcp
io.github.mikerawsonnz/authenticated-multi-llm-agent