Installation and usage guide#
This guide covers practical ways to install and run Metagit locally, plus how to enable project skills for AI-agent workflows.
Requirements#
- Python 3.12+
uvgit
Optional:
- Docker (for container usage)
gitleaks(for local security checks in QA workflows)
Option 1: global CLI install with uv (recommended)#
Install Metagit globally as a tool:
uv tool install metagit-cli
Upgrade later:
uv tool install -U metagit-cli
Verify:
metagit version
metagit --help
Option 2: install from source (local development)#
Clone and bootstrap:
git clone https://github.com/metagit-ai/metagit-cli.git
cd metagit-cli
uv sync
Install as an editable local package:
uv pip install -e .
Build a wheel and install that wheel:
task build
uv tool install dist/metagit-*-py3-none-any.whl
Option 3: container usage#
Run via container image:
docker pull ghcr.io/metagit-ai/metagit-cli:latest
docker run --rm ghcr.io/metagit-ai/metagit-cli:latest --help
First local use#
In a target Git repository:
metagit init --list-templates # bundled profiles (application, umbrella, hermes-orchestrator, …)
metagit init ./my-coordinator --template hermes-orchestrator --create
metagit init --target ../hermes-control-plane --template hermes-orchestrator --no-prompt \
--answers-file examples/hermes-orchestrator/answers.example.yml
metagit init --kind service --minimal # any ProjectKind without a bundled template
This creates .metagit.yml and updates .gitignore.
Useful first checks:
metagit config validate
metagit info
Local MCP runtime usage#
Start MCP runtime in current workspace:
metagit mcp serve
Run one-shot status snapshot:
metagit mcp serve --status-once
Pin a specific workspace root:
metagit mcp serve --root /path/to/workspace
AI-agent skill setup for this project#
If you cloned this repository and want skills available for local agent discovery:
task skills:sync
task generate:schema
One-shot combined:
task skills:sync generate:schema
This syncs project skills into .cursor/skills/ and regenerates config schemas.
Session closeout checks#
Before push or hand-off:
task skills:sync generate:schema
task qa:prepush
Troubleshooting#
- If
metagitfails after tool install, reinstall: uv tool install -U --reinstall metagit-cli- If command not found, ensure
uvtool bin path is onPATH. - If provider-backed features fail, configure API tokens in app config or environment:
METAGIT_GITHUB_API_TOKENMETAGIT_GITLAB_API_TOKEN