Everything in this manual so far happens in a chat with your assistant or in the admin. This chapter is for the other two places work gets done: a terminal, and the devices you record on. omumu-cli puts the whole tool surface of your site behind one command, for you, for a script, or for an agent such as Claude Code running on your laptop. Omumu Pod records a podcast episode on your phone and hands it to your site’s podcast pipeline. Omumu Loupe records your screen on a Mac, zooms and pans for you, and uploads the result into a lesson. The chapter says what each does, how it gets in, and where each stops — and which of them you can get today.
Concepts
One tool surface, several front doors. Your site’s AI connection (the Agent Surface chapter) is a set of named tools — omumu_offer_list, omumu_page_update and so on. The CLI calls those same tools over the same connection; it adds nothing the chat cannot do and leaves nothing out. What it adds is a terminal: typed arguments, JSON output, exit codes, and a self-describing schema an agent can read instead of guessing.
Two ways in. omumu login opens your browser and asks you to authorise the CLI on the site, exactly as you authorise a chat app — the Headmaster approves once, and the CLI gets an access token for that site. For a script or a scheduled job there is no browser, so you create an API key under Integrations → MCP API Keys with only the permissions the job needs and pass it with --token or the OMUMU_API_KEY variable. The browser login asks for a fixed, broad set of permissions and cannot ask for less; a narrow key is the way to give a script exactly what it needs.
The config file is a credential. Logins are stored in ~/.omumu/config.json, one profile per site (--name on login, OMUMU_SITE to choose one), readable by your user only. A browser login’s token lasts an hour and the CLI does not renew it; when a call says authentication failed, log in again. An API key lasts until you delete it.
Companion apps use the same sign-in. Pod and Loupe sign in to a site by its host name, in the system browser, with the same authorisation screen; their tokens live in the phone’s or Mac’s keychain and renew themselves. Each app then talks to the site through the tool surface: an upload, a tool call to create the episode or attach the video, and a poll until processing is done.
Hands-on
Install and log in
Install with Homebrew (brew tap omumuas/omumu && brew install omumu; a Mac on Apple silicon gets a native binary, other machines a jar that needs Java 21). Then:
omumu login --url https://mira.myomumu.comThe CLI registers itself with the site, prints the authorisation address and opens it in your browser; approve, and it reports Logged in. omumu status confirms the connection and counts the tools you can reach. Log in to a second site with --name to keep both.
Run any tool
Mira wants the week’s offers in a terminal: omumu call offer_list (the omumu_ prefix is optional). Every tool takes its arguments as --arg name=value, typed from the tool’s schema — numbers, booleans and JSON arrays or objects are parsed — or as one JSON object with --input, inline, from a file (@offer.json) or from stdin (-); use a file or stdin for anything secret, so it stays out of your shell history.
omumu call offer_create --arg title="Automate Your Week" --arg totalAmountCents=149000 \
--arg 'paymentProviders=["VIPPS"]'
omumu call page_update --input '{"pageId":"BBSR5F6MDX5OG","contents":"# About Mira"}'--json gives machine-readable output for every command, and omumu schema --json lists every tool with its parameters and the exact omumu call line for it, which is what an agent reads first. A misspelled tool or argument is refused with the known names. A handful of common tasks also have their own commands: omumu course list, get and create, and omumu skill upload for platform skills.
Let an agent drive it
Claude Code on Mira’s laptop gets a line in its instructions: “Omumu is managed with omumu call; run omumu schema --json to see the tools.” From there it lists, creates and updates like the chat assistant does, with one difference worth knowing: a terminal agent is only as careful as its instructions, and page and course writes are live at once. Give it an API key with the permissions the job needs rather than your browser login.
Record an episode on your phone
Omumu Pod, coming to Google Play, is the podcast chapter’s phone end: sign in to the site, record as many takes as you like (redo the last one with a tap), trim each on its waveform, reorder and merge them, then give the episode a title, show notes, a type and a season and number. The app uploads the audio, creates the episode, waits while the site masters it, and publishes it into the feed with one tap — or retries if mastering failed. The site needs a show first (the Podcasting chapter), and the Podcasting plan feature.
Record your screen on a Mac
Omumu Loupe is the Mac app we record our own tutorials with; it is not distributed to creators yet. It records a screen or a window with system audio and your microphone, follows your cursor and clicks with automatic zoom and pan, and gives you a timeline to adjust every move it made, with click ripples and keystroke badges if you want them. Export to MP4, MOV or GIF up to 4K — or sign in to your site and Upload to Omumu: pick a course, a module and the lesson the recording belongs to, or upload it as a standalone resource to place later (the Courses & Content chapter).
What you cannot do yet
Stated plainly: omumu signup, which creates a whole new site from the terminal behind a one-time card payment in the browser, is switched off on omumu.com today, so a site still starts at /trial (Getting Started). The CLI’s browser login asks for one broad set of permissions and leaves out the podcast tools, so podcast work from the terminal needs an API key that includes them; browser tokens expire after an hour and are not renewed. The CLI has no commands of its own beyond courses and skills — everything else is omumu call. Omumu Pod is not in the store yet, Loupe is not distributed, and Omumu Cam, a phone video recorder in the same family, is not released.
Reference
No MCP tools cover this chapter’s capabilities yet.