rimba mcp

Start a Model Context Protocol (MCP) server that exposes rimba’s worktree management as tools for AI coding agents. The server runs over stdio and follows the MCP specification.

All MCP tools return JSON responses. See rimba mcp help output for full parameter details. The easiest way to register this server is rimba init --agents, which writes the MCP config alongside the agent instruction files.

Synopsis

rimba mcp

Examples

rimba mcp    # Start MCP server (stdio transport)

Setup

Claude Code — add to .mcp.json or run:

claude mcp add rimba rimba mcp

Cursor — add to .cursor/mcp.json:

{
  "mcpServers": {
    "rimba": {
      "command": "rimba",
      "args": ["mcp"]
    }
  }
}

Auto-register with agent files:

rimba init --agents    # Writes agent files + MCP config simultaneously

MCP Tools

Tool	Description	Required Parameters
`list`	List all worktrees with branch, path, and status	—
`add`	Create a new worktree for a task (supports `service/task` for monorepos)	`task`
`remove`	Remove a worktree and optionally delete its branch	`task`
`status`	Show worktree dashboard with summary stats and age info	—
`sync`	Sync worktree(s) with the main branch via rebase or merge, then push	`task` or `all`
`merge`	Merge a worktree branch into main or another worktree	`source`
`exec`	Run a shell command across matching worktrees in parallel	`command`
`conflict-check`	Detect file overlaps between worktree branches	—
`clean`	Clean up stale references, merged branches, or stale worktrees	`mode` (prune, merged, stale)

Common workflows

Register once, use everywhere

rimba init --agents
# MCP config written to .mcp.json (Claude Code) and .cursor/mcp.json (Cursor)
# Agent instruction files written alongside

Ask your AI agent to create a worktree

"Create a worktree for the login-refactor task"
# Agent calls rimba MCP tool: add(task="login-refactor")

rimba init · --agents flag registers the MCP server automatically
rimba list · equivalent of the MCP list tool