Guide 26 · Workflow

Muse AI Connectors: The Beginner's Guide

Connect your calendar, inbox, and favorite apps — and know exactly what you're granting. A calm, complete walkthrough.

By Abhit Tomar · Updated November 16, 2026 · 15 min read

Minimal navy title card reading Muse AI Connectors Guide
Illustration: museaicodes

Short answer

Muse AI connectors link your agent to outside services — your calendar, email, or store accounts — so it can act on real information instead of guessing. Connect only what a task needs, keep approval checkpoints on, and review permissions regularly.

What a connector actually is

A connector is a permissioned link between Muse and one of your other services. Without connectors, Muse works from what you tell it; with them, it can check your real calendar before scheduling, read the actual email thread before drafting a reply, or look at live store listings before comparing prices.

Think of it this way: the chat is Muse's mouth, and connectors are its hands. You decide which hands it gets, one connection at a time — nothing is linked by default, and every link can be removed.

The difference between a connector and simply pasting information into a chat is that the connector stays current. A pasted calendar screenshot is stale the moment your schedule changes; a calendar connector lets the agent check the real calendar at the moment it acts. That is also why connectors deserve more care than chat history: a connection persists across conversations, so a permission you grant casually on day one is still live on day ninety. The good news is that every connector can be revoked the same way it was added, and reviewing them takes less time than the first setup did.

Availability of specific connectors depends on your account and region, and the roster grows over time — Meta has said more connectors and partners are on the way. The principles below hold no matter which services appear in your list.

The connections worth setting up first

Don't connect everything on day one. Start with the two or three services where Muse having real data changes the quality of its work:

  • Calendar — the highest-leverage connection. Scheduling, briefings, and conflict checks all get dramatically better when Muse sees your real calendar.
  • Email — worth it once you trust the agent: triage summaries, draft replies, and follow-up detection. Keep approval checkpoints on for anything sent.
  • Shopping and store accounts — price comparisons and purchase help stop being hypothetical when Muse can see real listings.
  • Files and cloud storage — summarizing your own documents and drafting inside your files beats copy-pasting every time.
  • Music and entertainment — lower stakes, great for learning how connectors feel: playlists, recommendations, and watchlists.

Business connectors and the 'find new customers' goal

Meta's small-business push gave Muse a second connector roster aimed at owners rather than individuals: Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe, and Zoom — alongside Facebook Pages, Instagram professional analytics, and Meta ad accounts. If a tool you rely on isn't listed, custom connectors are supported for services with a public API, at the cost of some manual setup.

The headline feature sitting on top of those links is the business goal. Tell Muse something like 'find new customers' and it turns the goal into a plan: drafting marketing content, preparing client pitches, building product pages — pausing for your approval at each step instead of publishing anything on its own. That approval-at-every-step behavior is the same safety model as the rest of the app, applied to business output.

Custom connectors deserve a sentence of their own, because they are the route for everything Meta has not listed. If a service you rely on has a public API, a connector can be built for it, but that is developer work rather than a toggle: someone has to write the integration, handle authentication, and keep it working when the service changes its API. For a small business, the practical question is whether the task is frequent enough to justify that setup, or whether the official roster already covers the tools that matter. Start with the official list and add a custom connector only for the one gap that actually costs you time every week.

Two practical notes before you wire up the whole stack. Muse is free for most uses, with paid tiers for heavy users who burn through large numbers of tokens — connect with a real workload in mind, not speculatively. And connector availability still varies by account and region, so treat the roster above as what Meta lists, not a promise of what your app shows today.

How connecting works, step by step

  1. Open the integrations or connections area in your Muse app or settings.
  2. Pick the service you want to link and read what access it requests — scopes matter more than the brand name.
  3. Sign in to that service through its official login page (never type credentials into a chat message).
  4. Confirm exactly what Muse may do: read-only is the safe default for anything you don't fully trust yet.
  5. Test with something small and reversible — 'summarize my inbox from today' — before handing it real responsibility.
  6. Tell Muse your preference once ('always ask before sending email') so it becomes a standing rule.

Permissions, approvals & staying safe

Connectors are powerful precisely because they touch your real accounts, so treat them with the same seriousness. The single most important habit: keep human approval on for anything hard to undo — sending messages, making purchases, deleting files, or changing settings. Muse's approval cards exist for exactly this; don't train yourself to tap 'approve' without reading.

Prefer read access where you can, grant write access only to services where Muse genuinely needs to act, and audit your connections every few months the way you'd audit app permissions on your phone. If a task is finished — the trip is booked, the project is done — disconnect what you no longer need. Our privacy guide covers the data side in more depth.

Never share passwords, one-time codes, or payment credentials inside a chat, even if asked. Legitimate connections happen through official sign-in screens, not conversation.

Connectors vs. channels vs. computer use

Muse reaches the outside world three ways, and it helps to keep them straight. Connectors are persistent, permissioned links to your accounts — set up once, used for months. Channels are where you talk to Muse: the app, WhatsApp, or voice mode. Computer use is the heavy machinery: Muse operating a real browser or desktop to complete bookings, forms, and purchases step by step, usually with you watching the approval checkpoints.

A good rule of thumb: connect for information (calendar, inbox, files), use channels for conversation, and reserve computer use for transactions. If you're curious about the heaviest of the three, the Mac computer-use guide explains what granting desktop access really means.

Beyond the official directory: community-built connectors

Everything above is about the connectors Meta ships and lists officially. There is also a second, unofficial world worth knowing about: an open community catalog of 150+ connector skills that independent developers build and share as open-source (MIT-licensed) code — things like GitHub, Slack, Notion, web-search APIs (Brave, Exa, Tavily), and Spotify, organised into roughly 30 categories.

These are not Meta products and they don't appear in the official connector directory. The model is different: instead of Meta vouching for a partner, you (or your agent setup) run a small open-source script that talks to a service's public API using your own API key or token for that service. Your key stays in secure credential storage and is only ever sent to that one service's official API host — the script itself never sees the real secret.

One more filter helps: look at whether the connector is actively maintained. A project with recent updates, a named maintainer, and issues that get answered is a different proposition from a script uploaded once and never touched. Check when it was last updated and whether it still matches the current version of the service's API, since an unmaintained connector can break silently and start returning stale or empty results. If you cannot tell from the repository whether anyone is still looking after it, treat the connector as an experiment, not infrastructure.

Hold them to a stricter standard before trusting one. Most community connectors are labelled draft by the community itself: written from the API's public documentation, not yet tested end-to-end against the live service (Slack is a rare exception marked as live-tested). Treat every one as a starting point, not a guarantee:

  1. Read the code before you run it. These are small, readable scripts — if you can't see what a connector does, don't give it a key.
  2. Check the license and the destination. A trustworthy connector names exactly one API host (for example api.github.com) and sends your credential nowhere else.
  3. Start read-only. List, search, and summarise first; let a connector create, post, or modify only after it has earned trust on reads.
  4. Make writes ask first. The better community connectors refuse to post, create, or save anything unless the exact action is re-confirmed — steal that rule for anything you build yourself.
  5. Keep keys out of chat. Add API keys and tokens through a secure credential flow, never by pasting them into a conversation.

Stuck? Troubleshoot, then keep exploring

Most connector problems are boring: an expired login, a revoked permission, or a service that changed its own API. Reconnect from the integrations screen, check the service's own 'connected apps' page, and try the smallest possible test before assuming something is broken. If a connector you expected isn't listed at all, it may simply not have reached your account or region yet.

Once your first two connections are working, the real unlock is combining them: 'find a free hour this week, draft the invite, and stop for my approval before sending.' That sentence uses a connector, a channel, and an approval card together — which is the whole agent idea in one line. For the patterns that make it sing, see how to use Muse AI and the prompt cheat sheet.

Keep exploring

How often should you audit your connectors?

Every few months, or whenever something changes. A finished project is the most common reason to disconnect: the trip is booked, the launch is done, the contractor has moved on, and the connection that served that project is still live, quietly holding permissions it no longer needs. App updates are the second trigger: a connector's scopes can change when either side ships an update, so a connection you set up as read-only can quietly gain new abilities. The audit itself is short. Open the integrations screen, read each connection's name and scope, and ask one question: would I still grant this today, knowing what I know now? Remove anything where the answer is no, and tighten anything where the answer is yes-but-narrower. Do the same check after any security scare, a password change, or a new device, since those are the moments when stale permissions matter most.

Can a connector see everything in my account?

No, and the limit is the point. A connector sees only what its scopes allow, and scopes are specific: read your calendar is different from edit your calendar, and both are different from managing your account. When you connect a service, the sign-in screen lists exactly what is being requested, and that list is the contract. Read it the way you would read a permission prompt on your phone, because it is the same thing. If a connector asks for more than the task needs, look for a narrower option or a read-only variant before accepting. You can also check the service's own connected-apps page at any time to see what Muse can currently reach, and revoke access there as well as in the Muse app. The habit that matters is treating scopes as a budget: spend the least access that still gets the job done.