On September 18, Mark Zuckerberg quietly opened a door that most founders haven't walked through yet.
Meta opened Muse, its personal AI agent, to outside developers. You can now build something called a connector, submit it for review, and have the number one app in the App Store recommend your service when someone asks for help.
Greg Isenberg watched the announcement and made a call: this could be the App Store moment for AI. He dropped a 26 minute video breaking down Meta Muse AI connectors, four startup ideas he'd build on top of them, and the exact way he'd ship a first version with a coding agent. It pulled 57,000+ views in a day.
Greg knows this terrain. He advised TikTok and Reddit, launched apps in the original App Store gold rush, and now builds AI-native businesses.
Here's the video, then my breakdown with the numbers that matter.
What Meta Actually Opened Up
A connector gives Muse a way to use your service mid-conversation.
Greg's example: he tells Muse he needs a podcast studio on a Tuesday afternoon, 2 hours, $200 budget. Muse needs a business that can answer whether a suitable room is free. Your connector is the service window it knocks on. The studio's system returns an available room at $160 with the gear included, Muse shows the offer, Greg approves, the booking gets made.
That's it. An API your software already has, exposed to the assistant that's sitting at number one in the App Store right now.
The submission side is real, not vaporware. Meta's platform page has a "Submit a connector" button, a three-step review (describe your product, pass functional, security and legal review with end-to-end testing by Meta, land in the directory), and Stripe Link handling payments. Editors can feature connectors. Onboarding is running in waves, and Meta hasn't published review times or fees yet.
Big companies are already in the directory. Greg's bet is that a wave of small and medium ones is next.
Why Greg Thinks This Is 2008 Again
When Apple opened the App Store in 2008, an outside developer could suddenly build for the phone Apple had already put in everyone's pocket. By June 2010, less than two years in, Apple had paid developers one billion dollars.
Greg lived that era. One editorial feature got his app 40,000 downloads a day. Free.
The same shape is forming here: Meta builds the assistant, outside businesses supply the services that complete a request, and a directory plus editorial featuring decides who gets discovered early.
There's already a serious company betting on it. Duffel, the travel API company, says its Muse integration lets people search flights and manage bookings inside the assistant. Duffel has passed one billion dollars in annual transaction value (that's bookings passing through the whole business, not revenue, to be fair). A company that size doesn't wire itself into a platform for fun.
And Greg points at a wrinkle the 2008 analogy doesn't have: your service can become relevant halfway through a bigger request. Someone organizing a small event starts by asking for a venue, then realizes they need to rent equipment. The rentals company enters the conversation at that exact moment, without the customer ever searching for it.
→ Watch for moments where someone is mid-task, needs a real business to deliver, and money changes hands. That's where connectors earn.
The 4 Startup Ideas Greg Would Build
Greg gave away four connector businesses. Here they are with his numbers.
1. Lead gen for business suppliers
A linen service in Miami supplies restaurants with tablecloths. It asks Muse: which restaurants are opening nearby that might need us?
Your connector answers from a service you built that tracks planned openings, with a source for each one and a public contact where available. General contact databases like Apollo already exist. Your edge is knowing something timely about one specific market.
His math: 100 customers at $99 a month is $9,900 MRR before costs. Start with one city and one supplier type. The hard part isn't the list, it's keeping it current and helping the supplier turn a fresh opening into an actual conversation (the follow-up side is a solved problem, these lead nurturing strategies cover it).
2. Home repair dispatch
"My dishwasher is broken. Find someone who repairs this exact Samsung model and can come tomorrow."
Your service checks participating repair companies and returns the ones with confirmed availability for that appliance. Charge the repair company an agreed fee, say $100, for a qualified intro or a confirmed booking.
Thumbtack already proved people pay for customer leads. The difference here is context: a confirmed opening tomorrow, delivered while someone is standing in front of a broken machine. Greg would start with one repair type in one town, because the repair companies need steady flow every day and the homeowner only shows up occasionally.
3. Local padel match and court finder
"Find me a padel game tonight near me with people around my level."
Your connector checks open spots at participating clubs, shows suitable games with prices, and books the reservation for a fee. Playtomic already does court booking, so Greg would hunt for the gap: local clubs whose availability players still have to check one by one.
He also expects Muse to get more social over the next 6 to 12 months. It's Meta. A club with an empty court has something to sell, and a player who plays three times a week has a reason to come back weekly.
4. Family dinner planning plus groceries
A parent says: sort out dinners for the next three nights, 20 minutes of cooking max, we already have rice and broccoli.
Your service builds the plan from tested recipes, adjusts quantities, and turns missing ingredients into a shoppable Instacart list the customer just reviews and checks out. Instacart's developer tools already support exactly that.
Greg's side note: some of these connectors probably get acquired by the platforms they plug into. If Muse gets big, an Instacart buying its best dinner-planning connector is not a crazy outcome.
Of the four, he'd start with the lead gen idea on a small budget. You can build a useful sample and show it to a paying customer before writing much software.
How People Will Actually Find Your Connector
This is the part most builders skip, and Greg spent a third of the video on it.
Getting into the directory gives you exposure, but nobody knows yet whether an unknown service gets recommended in a general conversation. So he'd build the launch plan around customers he can already reach:
- Creators with the right audience. For the dinner service, partner with a creator who shares vegetarian recipes for parents. Pay them for the customers they introduce. Audience size matters less than whether those people need the thing.
- Product-led sharing. One padel player books a court and shares a page with three other players. They get value before they ever sign up. Pick ideas with this multiplayer loop built in, the same pattern that made social apps win the mobile era.
- Existing marketplaces. Ticketmaster says eligible events can now appear through its connector with no extra work by the organizer. If your industry has a connected marketplace, riding it may beat building your own connector.
- Featured placement. The free 40,000 downloads a day outcome. Real, but you're banking on a product marketing manager in Menlo Park waking up and liking your app. Don't build the plan on it.
How Greg Would Build the First Version
In the App Store era, a high quality app took one or two million dollars and a team of ten. Greg's point: anyone watching can now build a connector solo.
His build path:
- Write down ONE thing the customer should accomplish. For padel: "show me available courts near me tomorrow evening, with prices."
- Hand that brief to Claude Code or Codex with the docs of the system you're connecting to. Ask for the availability check and the quote first. Booking comes after those work.
- Host it where Muse can reach it, and add auth so people only see what they're entitled to. Ask the coding agent to explain each piece as it builds (this is the same discipline as running a proper software factory: small pieces, verified, then shipped).
- Test the awkward requests before submitting: ask for a slot that's already booked, retry a request and make sure it doesn't double-book, move a booking to another day. Greg's blunt read: skip this and you get rejected.
Meta also documents a custom connector option, so you can wire your own Muse to your service and test a real integration before pursuing a public listing. The submission form (a founder at Stack Tree published a firsthand account) accepts an API or an existing MCP server, and asks for documentation, support details, and examples of real usage.
My Take
The honest unknowns: does Muse end up with 50 million users or a billion? And does Meta approve connectors like YC accepts founders, or like Apple approved apps, where most quality submissions got in?
Nobody knows. Greg's answer is the right one anyway: being early is cheap, and it buys you learning speed while your category sleeps.
This also picks up right where his 30-step AI SaaS playbook left off: agents become the default layer, and the money moves to whoever completes real tasks through them. Connectors are that, with Meta's distribution attached.
If I were picking, I'd take the lead gen idea, one niche, one city, and a spreadsheet-grade prototype shown to five suppliers before writing a line of code. Sell the sample first.
FAQ
What are Meta Muse AI connectors?
Connectors are integrations that let Meta's Muse assistant use an outside service to complete a request, like checking availability, returning a quote, or making a booking. Developers submit them at Meta's platform page through a three-step review, and approved connectors appear in a public directory where editors can feature them. Payments run through Stripe Link.
How do I submit a connector to Meta Muse?
Describe your product, submit it for Meta's functional, security and legal review (they test it end to end), and once approved it lands in the directory. The form accepts a standard API or an existing MCP server, plus documentation and support details. Onboarding is rolling out in waves, and Meta hasn't published review times or fees yet.
Do I need to be technical to build a Muse connector?
Less than you'd think. Greg's whole build path runs through a coding agent: write a one-sentence brief of the customer task, give Claude Code or Codex the docs for the system you're connecting, and have it build the availability check first. You still need to inspect what it built and test edge cases, or your submission gets rejected.
Is it too early to build on Muse?
Early is the point. Nobody knows if Muse ends up with 50 million or a billion users, but it's currently the number one app in the App Store, Duffel (over $1B in annual transaction value) already integrated, and the directory is thin. Early builders learn what customers actually ask for while the category is still empty.
Want the full playbooks from founders actually doing the numbers? Every week I sit down with bootstrapped SaaS founders making $100K to $10M a year and get the real story: pricing, distribution, the mistakes. No fluff.
