What Meta's Muse connectors mean for anyone building an AI startup right now

Ten days. That's how long it took Meta to go from launching its personal AI agent to handing developers a form and saying "build something for it." No SDK. No documentation dump. Just a one-screen page with a single button that says "Submit a connector."

That's not a small thing. It's the same move Apple made in 2008 when it opened the App Store — and within two years developers had pocketed a combined $1 billion from that decision. Meta just made the equivalent move for AI agents, and almost nobody outside a small circle of builders has noticed yet.

This guide breaks down what a Muse connector actually is, how the submission process works right now, and what kind of business you could realistically build on top of it before the directory fills up.

What is a Muse connector, exactly?

A connector is the bridge between Meta's AI agent and your service. Per Meta's own help documentation, a connector links Muse to an app or service so the agent gains an ability, such as reading a calendar or writing an email. Instead of a user opening your app, they just ask Muse for what they want, and Muse decides which connected service can fulfill that request.

Alexandr Wang, Meta's Chief AI Officer, described the pitch bluntly when announcing the program: we're opening muse connectors to developers, and we have seen so much excitement in the developer community, integrating muse into everything from robots to mood lights and more. The instruction to developers is simple — plug your API into muse, then people can use your service just by asking for it.

There are actually two flavors of connector, and the distinction matters a lot if you're planning to build one:

Directory connectors. These get reviewed by Meta and show up in Muse's built-in connector list, which any user can browse and enable. Two classes exist: connectors listed in the directory and reviewed by Meta, and custom connectors Muse builds for one user from a service's API, which Meta does not review.

Custom connectors. These are one-off integrations Muse builds on the fly when a user asks it to connect to something not yet in the directory. According to Meta's help center, if you want to connect to a service not yet available in the Connector list, you can ask Muse to create a Custom Connector. Wang has even said publicly that you can often just ask muse to connect to anything and it'll write the whole integration in the background.

That second point is wild if you sit with it. It means Muse can, in theory, wire itself up to your public API without you doing anything at all — good or bad, depending on whether your API is ready for that kind of traffic.

How does the submission process actually work right now?

Right now, Meta's connector submission process runs through muse.ai/platform, and it's refreshingly bare-bones. There's no SDK to download and no lengthy developer terms to sign — at least not yet. As one early submitter described it, a connector is either an existing MCP server or a raw API, submitted through a three-step form on muse.ai/platform: describe the product, give the technical details, confirm and submit.

Once you submit, Meta puts your connector through a review process. The platform page states that Meta reviews submissions for functional, security and legal requirements and completes end to end testing, and that Meta has partnered with Stripe so connectors can accept payments with Link. If you're planning to charge for anything through your connector, that Stripe Link partnership is worth paying attention to — it suggests Meta wants transactional connectors, not just read-only integrations, to be a real part of this ecosystem.

Here's what to expect on timing, based on what Meta has told early submitters. A follow-up email sent to developers about a week after the form went live confirmed over 2,000 submissions in the first few days, a developer portal to track status "in the coming days", and onboarding "in waves over the coming weeks". Translation: this is not an instant-approval situation. If you submit today, expect to wait weeks rather than days, and expect the process to keep evolving as Meta figures out how to handle the volume.

A few practical notes before you start building:

Go to muse.ai/platform and click "Submit a connector." That's currently the entire front door — no separate developer console to sign into first.
Have your API or MCP server ready. The form asks for a product description and technical details, so you'll want your API endpoint documented and stable before you submit, not half-built.
Think through the security and legal side early. Since Meta explicitly reviews for security and legal requirements, anything that touches user data, payments, or health information should be locked down before submission, not patched afterward.
Check Meta's Muse help center for the latest on what's supported, since the built-in connector list and rules are still shifting week to week.

Why does this feel like the app store moment all over again?

The App Store comparison isn't just a marketing hook — it's structurally similar. Before 2008, building a mobile product meant convincing carriers to bundle your software or building for the mobile web, which was clunky and barely discoverable. The App Store created a single distribution channel where a small team could reach millions of phones directly.

Muse connectors do something similar for AI agents. Instead of building a standalone app that a user has to discover, download, and open, you build a connector that plugs into a conversation someone is already having. The user doesn't go looking for your service — Muse decides your service is the right fit for the task and reaches for it mid-conversation.

And the audience Muse already has isn't small. Muse, powered by Muse Spark 1.3, already demonstrates rapid adoption, hitting 642,000 US mobile daily active users in just 12 days—three times ChatGPT's initial pace. A few days later, momentum kept building — Muse quickly jumped to the top of both Apple's App Store and Google's Play Store. That's the audience your connector could potentially reach, without spending a dollar on user acquisition, if Meta's directory surfaces it for the right requests.

What kinds of businesses actually fit this model?

Not every business idea translates well into a connector. The ones that work best share a pattern: someone describes a need in plain language, and fulfilling that need requires checking availability, price, or booking status from a real business in real time.

A few categories worth considering:

Local service dispatch. Think home repair, plumbing, or appliance fixes. Someone tells Muse their fridge broke, and a connector checks which nearby providers have availability today. The repair company gets a steady flow of leads instead of relying on word of mouth or expensive local ads.

Booking and availability checkers. Restaurants, studios, courts, or rooms for hire all fit this pattern well, since the core job is answering a yes/no question about a time slot. This is also the exact use case Meta itself has already built in with connectors like OpenTable, per Wang's launch-day list of built-in integrations, which named Gmail, Google Calendar, Outlook, Plaid, OpenTable, Google Docs, Spotify, Function Health, Withings, Tailscale and Peloton among the first connectors.

Niche lead generation. If you sit between a supplier and a business — say, matching event venues with caterers — a connector lets Muse do the matching conversationally instead of someone filling out a five-field lead form on a website.

Planning and coordination tools. Family dinner planning, group trip coordination, or matching people for recurring activities like a paddle match all fit a similar shape: Muse gathers context across a conversation, then needs a service that can act on that context.

The common thread: pick a task where the request itself is the hard part to automate, and the fulfillment step is something your business already does well.

How do you build and test a first version fast?

You don't need a ten-person engineering team to get a connector working today. A coding agent like Claude Code or OpenAI's Codex can scaffold an API or MCP server for you in an afternoon, especially if your core logic is simple — check availability, return a price, confirm a booking.

The workflow that makes sense here: build the smallest possible API that answers one specific question well, wire it up as an MCP server (since that's one of the two formats the Muse platform form accepts), and test it yourself by asking Muse the exact phrasing a real customer would use. Don't just test the happy path — try the awkward, oddly-worded requests too, since that's where most connector demos actually fall apart.

Before you submit to Meta's review queue, get a handful of real users testing the flow first. Start with one type of customer, watch whether they complete the task without help, and only then worry about scaling the audience. A connector that works for ten people who genuinely needed it beats one that technically works for ten thousand who never use it twice.

What should you do this week?

If you've been sitting on a service idea that needs "an app" to work, this is the moment to reconsider that assumption entirely. You might not need an app. You might just need an API and a connector.

Head to muse.ai/platform, read through the three-step submission flow, and start sketching what your connector would answer when someone asks Muse for help. The directory isn't crowded yet. Every app store had an early window where getting listed was easy and the competition was thin — this is that window, and it's open right now.