# Agents and the standalone CLI





This page covers the command-line `emm` tool: getting it on PATH, finding help, reading an issue, and giving an agent its workflow.

## Get the CLI on PATH

The `emm` command-line tool is bundled with the Emm macOS app in `Emm.app/Contents/Helpers`. In Emm, go to **Settings → Command-line tools** and choose **Add to PATH**, then open a new terminal so your shell picks it up. The prompt names the directory and shell profile; accepting it is required — merely opening Settings does not change PATH.

Alternatively, download the standalone `emm` CLI for macOS or Linux from the [CLI download page](https://emmflow.com/cli/). The Markdown tool `mdt` is versioned independently, so its download page choice is separate from Emm's.

## Find help

In a terminal inside your project folder:

```sh
emm --help
emm issue --help
```

Drill into a subcommand with `emm <subcommand> --help`, for example `emm issue get --help`.

## Read a whole discussion

In a terminal inside your project folder, use a parent message or reply UUID to read the parent and every reply:

```sh
emm thread get --message MESSAGE_UUID
```

Replace `MESSAGE_UUID` with the message’s UUID. If you already have the thread UUID, use `emm thread get THREAD_UUID`. The parent appears first; replies follow in timestamp and UUID order without a result limit. Reading a parent with no replies returns that message without creating a thread. Access to private conversations follows the project’s current identity.

This reads your local project copy. Run `emm sync` first if you need fresh shared content. Use `emm thread get --help` for output details and errors.

## Read an issue

List issues to find a number, then read that issue:

```sh
emm issue list
emm issue get 42
```

`42` is the issue number; replace it with a real number from your project. IDs also accept `#42`, a UUID, or a unique UUID prefix.

## Give an agent its workflow

`emm agent prompt` prints the opinionated workflow for an agent working the queue, with this executable and the selected project's store path. Run it at session start and paste the output into an agent's instructions. It reads paths only; it does not claim issues or start work.