Install and update M3

Install the M3 CLI and the project SDK separately. The CLI runs m3 and includes the local results viewer. The SDK runs inside the Python project being tested.

Requirements

M3 supports Python 3.10 and newer. The commands below use uv. The alternative shell installer supports macOS and Linux and requires curl plus either Python 3.10 or uv.

Install the CLI

Install the standalone command once for your user with uv:

uv tool install sf-m3-cli

The CLI runs outside your project environment. To update it, use:

uv tool upgrade sf-m3-cli

After uv tool upgrade sf-m3-cli, rerun m3 setup to upgrade the project SDK.

On macOS or Linux, you can use the shell installer instead:

curl -LsSf https://m3.sineframe.com/install.sh | sh

The installer selects the latest stable GitHub release, verifies its release installer and checksums, and installs the CLI in isolated tool storage. It uses uv when available and otherwise creates a dedicated virtual environment. To update an installation made this way, rerun the same command, then rerun m3 setup in each project to upgrade its SDK.

Install the project SDK

From a project directory, initialize M3 and prepare its Python environment:

m3 init
m3 setup
m3 doctor

m3 init creates the project identity and a skipped pytest starter. Replace that starter with a real assertion before relying on test results. m3 setup installs the SDK release matching the CLI, with pytest, storage, and judge support, into the selected project environment. It does not install the CLI in that environment or edit dependency manifests and lockfiles. m3 doctor checks the CLI and project environment separately.

Keep the CLI and project SDK on matching releases. If the environment is recreated or synchronized, run m3 setup again. For a project that manages its dependencies directly, add sf-m3[pytest,storage,judge] as a project dependency instead of using CLI-managed setup; you still install the standalone CLI separately when you want m3 test, saved history, or the bundled viewer. m3 test checks that the project environment can import pytest support, SQLite storage, and the openai package used by LLMJudge; sf-m3[pytest] alone is enough only for plain pytest without the CLI.

With uv, add the SDK from the project root with:

uv add "sf-m3[pytest,storage,judge]"

This declares the SDK in the project manifest. The standalone CLI remains a separate user-level installation.

To upgrade the SDK in a project that manages its dependencies directly, run uv lock --upgrade-package sf-m3 (or the equivalent for your tool), then sync the environment. You do not need m3 setup for this setup.

Try a canary build

Canary builds let you try unreleased changes before they ship. They are published as GitHub prereleases, never to PyPI, and are not supported releases. canary-main follows the main branch, and canary-pr-N follows pull request N while it has the canary label.

# Latest build of main
curl -LsSf https://m3.sineframe.com/install.sh | sh -s -- --canary

# A pull request
curl -LsSf https://m3.sineframe.com/install.sh | sh -s -- --pr 123

Each canary's release page, such as canary-main, shows its version and the exact install commands, including a pinned install.sh link and a uv tool install command. m3 --version and m3 doctor name the installed canary and its commit. m3 setup installs the matching SDK from the same canary release. To return to the latest stable release, rerun the installer without options.

Continue to your first MCP test, or test your own server.