Claude Code Agent Notification App: Muster for Your Mac

Why Agent Fleets Create a Notification Problem

Running several coding agents at once sounds great until you hit the waiting problem. You start a Claude Code session to fix a bug, open a second one to write a test, and let a Codex session churn on a refactor. Ten minutes later, all three are sitting at a prompt waiting for permission to run a command or waiting for you to answer a question. You are in another window. Nobody moves until you notice.

If you are using a Claude Code agent notification app, that gap should close. Muster is a free, open-source Mac menu bar app that watches your Claude Code, Codex and Cursor sessions, shows you which one needs attention, and lets you answer with one keystroke. It is aimed at people who run agent fleets, and that includes anyone using Creator OS to post, reply and read analytics with an AI agent on top.

  +----------------+   +----------------+   +----------------+
  | Claude Code #1 |   | Claude Code #2 |   | Codex session  |
  | "run migrations|   | "create post?" |   | "refactor test |
  |  ? allow/deny" |   |  ? y/n/edit    |   |  ? allow/deny" |
  +--------+-------+   +--------+-------+   +--------+-------+
           |                    |                    |
           v                    v                    v
  +-----------------------------------------------------------+
  |   Muster (Mac menu bar)  ->  pick the session waiting   |
  |   one keystroke = one answer, no window switching       |
  +-----------------------------------------------------------+
           |
           v
  Your Creator OS agent runs create_post, list_comments,
  get_post_analytics ... and you approve before publish.

Muster surfaces the agent that needs you and clears it with one key.

Get started for $0

What Muster Does: A Menu Bar for Your Agent Sessions

The pitch from the project is blunt: never keep an agent waiting. Muster takes the status of each session, puts it in the Mac menu bar, and gives you a small surface to act on it. If a session is blocked on a permission prompt, Muster shows that. If three sessions are waiting, you see which ones and clear them in order.

The keystroke part is what makes it useful. A permission prompt is usually a trivial yes or no, and clicking back into a terminal window to press a key is friction. If the answer is one keypress away from wherever you already are, you stop being the bottleneck. That matters most when you are running more than two agents, because the cost of context switching is what kills throughput.

It is useful to be clear about what Muster is not. It is not a model, not a hosted service, and not a replacement for Claude Code or Codex. It is a control surface in front of tools you already run.

How Muster Fits Claude Code, Codex and Cursor

Many people run one editor and one terminal agent. Muster’s value shows up when you mix them. A typical setup might be:

Cursor open in one window editing files.
A Claude Code session running a task in a terminal.
A second Claude Code session handling something unrelated.
Codex doing a longer job you do not want to supervise.

Each of those can stall on an approval. Without a single place to look, you end up cycling through windows to find whoever is waiting. Muster gives you one place to look. The mix also fits Creator OS work cleanly, because the same pattern shows up in content ops: one agent drafts a post, another reads comments, a third pulls analytics.

If you want the setup for that broader agent workflow, the walkthroughs on the KevBuildsApps YouTube channel cover agent builds and the open-source tooling around them. The launch video at youtube.com/watch?v=-QBJH_PK3pY shows the open-source pieces in action.

Install Muster on Your Mac in a Few Minutes

The source and install instructions live at github.com/josefchen/muster. Read the README there before you install; it is the authoritative place for build steps, permissions and any setup Muster needs on your machine. Do not guess at the commands. Follow what the repo says for your version.

Once it is running, expect two things to matter:

macOS permissions. A menu bar app that watches other apps will trigger system prompts. Grant what the README asks for, nothing more.
Session visibility. Start your Claude Code, Codex or Cursor sessions the way you normally do, then confirm Muster sees them. If a session does not appear, that is the first thing to debug.

Before you install, think about scope. Muster shows session state, which for an agent may include command text and prompts. Treat that as sensitive if your sessions touch client credentials or API keys.

Setting Up a One-Keystroke Approval Step Before Posts Go Live

Here is the workflow this really unlocks. Creator OS ships a hosted MCP server and a CLI, so an AI agent can create posts, schedule them, reply to comments and read analytics. The MCP server is at https://mcp.creatoros.ca/mcp and nothing needs installing. For Claude Code you add it with:

claude mcp add --transport http creatoros https://mcp.creatoros.ca/mcp \
  --header "Authorization: Bearer cos_live_..."

The tool list includes create_post, update_post, list_comments, reply_to_comment, get_post_analytics and skool_create_post, among others. Each is a step your agent can take. Destructive tools like delete post, delete comment, disconnect account and delete ad are marked so the AI app asks first.

Now put Muster in front of that loop. A session drafting and publishing looks like this:

You tell Claude Code: draft three posts for this week, check caption length, then queue them as drafts.
The agent calls check_caption_length, then create_post with drafts.
Muster surfaces the session when it wants to go from drafts to scheduled or published.
You review the draft, answer with one keystroke, and the agent proceeds. Or you say no and it revises.

That is a human-in-the-loop gate that costs you a keypress instead of a window switch. It is the same idea in the post on letting an AI agent post to Instagram: the agent does the work, a human clears the final step.

Running Muster Alongside Your Creator OS Content Agents

Creator OS works from the web app, the iOS app, a REST API, the CLI (@creatoros/cli) and the hosted MCP server. Every plan includes API keys, the MCP server, the CLI and the agent skills. If you want a starting point for the agent itself, the guide on building a social media agent with Claude Code covers it, and npx @creatoros/cli@latest init installs 14 Claude Code skills including post-everywhere, respond-to-comments and analytics.

A three-terminal setup that pairs well with Muster:

# terminal 1: draft and schedule
creatoros posts:create --text "New tutorial is live" \
  --platforms instagram,tiktok --media med_123 \
  --scheduledAt 2026-03-02T15:00:00Z --timezone America/Toronto

# terminal 2: comment triage
creatoros inbox:comments --platform instagram --limit 20

# terminal 3: weekly numbers
creatoros analytics:posts --sortBy engagement --limit 10

Three sessions, three potential approval prompts, one menu bar. Each command is something you could also give to an agent through the CLI or MCP. When the agent asks whether to send a reply in your voice, Muster is where you catch it. The post on replying to Instagram comments in your voice is the deeper read on that loop.

If you also run a Skool community, an agent can create posts, moderate members and read revenue through skool_create_post and related tools. Those are exactly the calls you want gated, since a bad post to a paid community is public and awkward. More at the Skool docs.

Limits, Security and When Not to Auto-Approve

Three practical rules:

Do not blanket-approve writes. A keystroke is a decision. Read the prompt before you press it, especially for anything that publishes, deletes or changes budgets.
Keep read-only connections where they belong. Creator OS MCP connections can be read only or read and write. A read-only connection only sees read tools and the API refuses writes from it. If an agent only needs get_post_analytics and get_best_time_to_post, give it read only.
Scope API keys by workspace. Keys are pinned to one workspace, so an agent holding one workspace’s key cannot post to another.

On ads, Creator OS ads can be dry-run with validate_only, budgets are confirmed, and ads are created paused for review. Add Muster on top and the “are you sure” moment lands in your menu bar instead of a terminal you forgot about. The ads docs cover the details, including the Meta cap of about 30 creates per ad account per 5 minutes.

Where Muster is a poor fit: fully unattended jobs where no human is present, and long batch runs where a prompt every few seconds is worse than no prompt. For those, configure the agent to run within limits it can hold without asking.

Get Started: Wire Muster Into Your Creator OS Workflow

Install Muster from the repo, connect Creator OS’s MCP server or install the CLI, and run one agent session end to end with Muster watching. Start with drafts so the first approval is low risk. The MCP docs and CLI docs have the connection details, and creatoros.ca/docs is the index for everything else.

The Creator plan is $14.99/month or $59.99/year for 11 platforms plus Skool, one account on each. API keys, the MCP server, the CLI and the agent skills are included. Sign up at creatoros.ca/sign-up.

If you would rather watch a build than read one, the KevBuildsApps YouTube channel has the walkthroughs.

Get started for $0
Free trial for new subscribers. Cancel anytime.