Where does my meeting data live?

Local SQLite by default, with Supabase sync limited to your account. No one else can access it.

Can participants tell they're being recorded?

Bot-free capture — nothing visible by default. Recording notice is available as an opt-in setting.

When is Windows coming?

It's here — grab the Windows 10 · 11 (x64) build from the download page. Auto-detection covers Zoom only for now; start Teams and Google Meet recordings yourself with the record button.

Can I import my Otter · Fireflies · Tiro data?

Yes — transcript import is supported. Existing meetings feed the Team Agent immediately.

How accurate is it for non-English meetings?

Whisper Large + domain keyword tuning. Teams can register their own vocabulary and titles.

← Home

Docs · Developers

Micky CLI · MCP

The official guide to using your meeting knowledge from the terminal and AI agents. From install to Claude Code in about five minutes.

Overview

There are two ways to use it. Most people connect Micky to their own AI (Claude Code, Cursor, Codex, Gemini) and just ask in plain language — no commands to memorize, like “what did we decide about this last meeting?” Typing commands in the terminal is a separate path, for scripts and automation. Install, log in, then connect your AI below — a few minutes total.

Prerequisites

Node.js 18+ (includes npm)
A Micky account — read commands with login; AI generation (ask · digest) on beta-approved accounts
macOS · Linux · Windows (WSL recommended)

1. Install

Install globally with npm. No registry login needed — it pulls straight from GitHub Releases.

npm install -g https://github.com/Marco-Joshua/micky-cli-releases/releases/download/v0.1.3/micky-ai-0.1.3.tgz

Once installed, verify with:

micky --version

Tip

If a corporate proxy blocks the install, download the .tgz from the URL above and run `npm i -g ./micky-ai-0.1.3.tgz`.

2. Log in

A browser opens; sign in with Google and the session is stored securely for the CLI (~/.micky). One time only.

micky login

If the browser doesn't open, paste the URL shown in the terminal. Check your session with `micky whoami`.

3. Connect your AI · ask in plain language (MCP)

This is the main way to use Micky. Turn on MCP (Model Context Protocol) and your AI sees your meetings as tools. Instead of commands, just talk to it normally — ask “what did we decide about this last meeting?” and it finds the meeting and answers on its own.

micky mcp install claude

Other MCP clients like Cursor: paste the config snippet the command prints into their MCP settings.

Tools your agent gains

semantic_search — find meetings by meaning (free)
create_digest — combine meetings into one artifact
ask_micky — ask across meeting knowledge
list_meetings · get_meeting · get_transcript — read meetings
list_action_items — open actions across meetings
open_loops — open commitments bucketed by urgency
get_usage — check usage

Now just tell your AI things like

“Pull together just my action items from this sprint's meetings.”

“Find what auth method we settled on in the API design meeting.”

“Implement this payment function per the last planning meeting.”

“Summarize our meetings with Client A and draft a brief for the next one.”

“List the promises from recent meetings that aren't done yet.”

“Turn last week's meetings into a Markdown recap for the team.”

Use cases

In the terminal, and inside your AI agent — here's what it looks like in practice.

AI agent

You need meeting context mid-code

"What auth method did we settle on in the API design meeting?"

Claude Code / Cursor finds that meeting and answers — no digging through Slack, right in your flow.

Terminal

Weekly team roundup

micky digest <this week's ids> --mode actions

Every meeting's to-dos on one page, grouped by owner. Cron it and get it every week.

Terminal

A 5-second brief before a call

micky search "Company A on-prem terms" --semantic

Surfaces the meeting where it came up — free, and by meaning even when the words differ.

AI agent

Delegate work per a decision

"Implement this per the payment flow we agreed in planning"

The agent references the decision and writes the code — no re-explaining the spec.

Terminal commands (advanced)

You don't need to memorize commands — connecting your AI above and asking in plain language is the default. Use these for scripts, CI, cron, or when you prefer the terminal. Add `--json` to any command for script-friendly JSON.

Command	What it does	Cost
micky meetings	List your recent meetings	Free
micky meeting <id>	One meeting: metadata + summary + decisions + actions	Free
micky transcript <id>	Full transcript text	Free
micky search "<query>"	Keyword search over titles and summaries	Free
micky search "<query>" --semantic	Rank meetings by meaning (embeddings) — context, not keywords	Free
micky digest <id...> --mode <mode>	Combine meetings (summary · actions · decisions · email)	LLM
micky ask "<question>"	Ask across all your meeting knowledge (RAG)	LLM
micky actions	Open action items across all meetings	Free
micky loops	Open commitments bucketed by urgency (overdue · dueSoon · upcoming · undated)	Free
micky export <id> -o note.md	Export a meeting to Markdown	Free
micky usage	Credits & recording usage	Free

semantic search

`--semantic` finds meetings by meaning via embeddings — searching “billing reconciliation” surfaces the “month-end close” meeting. No LLM synthesis, so it's free.

digest

`--mode` is one of summary (default) · actions · decisions · email. Pass meeting ids space-separated (up to 60) from `micky meetings` or `search`.

Troubleshooting

ask / digest returns a 402

AI-generating commands run on beta-approved accounts. Check status with micky usage. Read commands (search · transcript · export) work with just login.

--semantic returns nothing

The meetings may not be embedded yet. They're indexed automatically shortly after a meeting ends — wait a moment and retry.

command not found

Confirm the install finished and open a fresh terminal. Get the latest version from Releases below.

Next steps

GitHub Releases (latest version) →