Use via MCP

axm-git exposes its CLI commands as MCP (Model Context Protocol) tools via axm-mcp. AI agents can call them directly without spawning subprocesses.

Setup

These tools are served by axm-mcp. If you haven't connected the server yet, see the axm-mcp Quick Start — one command connects the whole toolchain. No per-package install needed.

Available Tools

MCP Tool	Purpose
git_preflight	Working tree status and diff summary before a phase
git_branch	Create or checkout a branch
git_commit	Batched atomic commits with pre-commit; warns on non-Conventional-Commit messages (strict=True blocks them), retries on pre-commit auto-fixes
git_clone	Clone a repository into a local directory
git_tag	One-shot semver tagging (skips CI checks when gh is unavailable)
git_push	Push with dirty-check and auto-upstream; force uses --force-with-lease by default
git_pull	Pull origin main (override via remote / branch) into the local repo
git_worktree	Add, remove, or list git worktrees
git_pr	Create GitHub pull requests with optional auto-merge; idempotent — recovers an existing open PR
git_merge	Squash-merge a branch back into its target
git_await_merge	Poll a PR (pr_number / pr_url) until merged or timeout

Usage

MCP dispatch

The examples below show the logical API — the parameters each tool takes. In practice, AI agents call these via MCP tool dispatch (e.g. mcp_axm-mcp_git_commit), not direct Python imports.

Text Only

git_preflight(path="/path/to/repo")
git_branch(name="feat/new-thing", path="/path/to/repo")
git_commit(path="/path/to/repo", commits=[{"files": ["src/foo.py"], "message": "feat: add foo"}])

git_commit handles staging internally (including deletions via git add -A), so you never run git add separately.

Entry Points

All tools are auto-discovered via the axm.tools entry-point group — see the CLI Reference for the full tool / class mapping. axm-mcp discovers these automatically at startup.