TreeMCP
MCP server for git worktree lifecycle — give every parallel AI agent its own isolated sandbox in one tool call
npx -y github:Isco0819/tree-mcpEver run into this?
Like git worktree (built into git, used by millions for parallel branch work) but the UX is broken — you juggle absolute paths in your head, forget to prune stale trees, and there is zero structured interface for AI agents. Teams running multiple parallel agents on the same repo manually track worktree paths in sticky notes.
The fix
An MCP server that creates, lists, and garbage-collects git worktrees on demand. Agents call worktree_create to spin up an isolated branch sandbox and worktree_release when done — no path juggling, no leaked dirty trees.
What it does
- worktree_create(branch) — checks out a branch into a fresh worktree and returns the absolute path ready for the agent to operate in
- worktree_list — returns structured JSON of all active worktrees with branch name, dirty flag, and idle-since timestamp
- worktree_cleanup — prunes merged and stale worktrees automatically and reports exactly what was removed