Meta has opened a new distribution surface for software businesses: approved third-party services can become connectors inside Muse, its personal AI agent. Instead of asking a user to discover an app, learn its interface, and repeat their context, a connector can appear inside a task the user is already trying to finish.
Watch Greg Isenberg's Full Muse Connector Episode
Source and credit: Meta Muse AI Connectors: The Next App Store Moment?, published by Greg Isenberg and the Startup Ideas Podcast on 24 September 2026. The four startup concepts, growth routes, and prototype brief come from Greg's episode. Platform facts are checked against Meta and the named companies.
What Meta Actually Confirms
Meta's public Muse Connector Platform page describes a simple three-stage path: describe the product, submit it for review, and appear in the directory after approval. Meta says the review covers functional, security, and legal requirements and includes end-to-end testing. Approved connectors may then be considered by editors for featured placement.
| Platform fact | Evidence status | Builder implication |
|---|---|---|
| Outside developers can submit connectors | Official Meta platform page | The channel is open, but approval is not guaranteed |
| Functional, security, and legal review | Official Meta platform page | A demo endpoint is not enough; behavior, policies, and support must survive testing |
| Directory listing after approval | Official Meta platform page | Discovery exists, but Meta publishes no traffic guarantee |
| Editorial consideration for featured placement | Official Meta platform page | Treat a feature as upside, not the acquisition plan |
| Stripe Link supports payments | Official Meta and Stripe material | Purchase approval and transaction infrastructure are part of the platform story |
| Raw API or hosted MCP submission | Stacktree's firsthand form walkthrough | Verify the live form; detailed technical documentation is still limited |
Stacktree's founder documented the form after submitting on 19 September. It reportedly asks for a live raw API or existing hosted MCP endpoint, documentation, authentication details, example prompts, product and support information, privacy and terms links, and payment status. This is useful firsthand evidence, not a substitute for Meta's current form or future developer documentation.
Why the App Store Analogy Is Useful, and Where It Breaks
The analogy works because Meta supplies the agent, user relationship, permissions, directory, and payment path while outside businesses supply specialized services. Apple created a comparable distribution shift when it opened the App Store. Apple says developer payouts from paid apps and in-app purchases reached $1 billion by June 2010.
Muse changes the discovery moment. A connector can become useful halfway through a larger request: equipment rental while planning an event, a repair provider after diagnosing a broken appliance, or a court booking while organizing a game. The business opportunity sits where the conversation reaches a step that requires live supply, reliable execution, and money changing hands.
Two launch integrations show the model. Duffel gives Muse live flight search, booking, management, and cancellation infrastructure across more than 500 airlines. Ticketmaster exposes eligible live-event inventory while keeping final purchases on its own marketplace. These are existing operational systems gaining a new interface, not businesses created from a connector alone.
Four Muse Connector Business Wedges
| Idea | First narrow market | Value event | Hard dependency | Possible model |
|---|---|---|---|---|
| Business-supplier lead intelligence | Restaurant openings for one supplier type in one city | A verified, timely prospect with evidence and a public contact route | Licensed sources, freshness checks, and human verification | Subscription or paid qualified list |
| Home-repair dispatch | One urgent repair category in one service area | A provider accepts a job they can actually serve | Live availability, coverage, qualification, cancellation, and support | Qualified introduction or confirmed-booking fee |
| Padel match and court finder | Independent clubs in one city | A suitable game or court is reserved at a confirmed price | Club permission, live inventory, stable quotes, and booking recovery | Booking commission or club software fee |
| Family meal and grocery planner | One recurring dietary or household profile | A useful meal plan becomes a reviewed shopping list | Tested recipes, substitutions, family feedback, and grocery availability | Subscription, affiliate revenue, or commerce partnership |
Greg would investigate supplier lead generation first on a small budget because a founder can assemble and sell a useful sample before building much software. That is a strong selection rule: choose the opportunity whose value can be demonstrated manually, then automate the repeated parts after someone pays.
The meal-planning example has a credible fulfillment path through the Instacart Developer Platform, which supports meal planning, shoppable recipes, catalogs, and shopping lists. Access and permitted use still depend on Instacart's current program requirements.
Do Not Make Meta's Directory Your Entire Growth Plan
- Partner with a trusted niche creator. Demonstrate one complete task to an audience that already experiences the problem. Measure completed first jobs, not views.
- Build sharing into the result. A court reservation, event proposal, or family plan naturally reaches other people. Make the shared output useful before asking the recipient to register.
- Enter through an existing marketplace. Ticketmaster says most on-sale events with active inventory can be eligible for Muse discovery without each organizer building an integration. In other sectors, the practical route may be supplying data to an established connector.
- Earn direct customer demand. Reach one buyer group through sales, partnerships, community, or content. The connector should improve conversion and convenience, not hold the only relationship.
A directory feature can create a surge. It cannot be forecast. The healthier plan names the first ten customers and the channel that can reach them before Meta's recommendation system matters.
Build One Complete Outcome First
For a padel prototype, the first outcome is not "build a booking platform." It is: show available courts near a location for a specific date, start time, and duration, then return a complete quote. Use fictional, clearly labeled inventory until supplier access is authorized.
| Stage | What to build | Acceptance test |
|---|---|---|
| 1. Search | Location, date, start time, duration, and clearly labeled sample clubs | Unavailable and malformed requests return useful errors |
| 2. Quote | Total price, currency, expiry, assumptions, and stable quote ID | The same request is reproducible and an expired quote is rejected |
| 3. Approval | A visible confirmation step before any reservation | No write action occurs from a search or quote request |
| 4. Reservation | Sandbox booking with an idempotency key | Repeating the same approved request does not create a duplicate |
| 5. State | Reservation lookup, status, and audit record | A reviewer can verify exactly what changed |
| 6. Delivery | Reachable HTTPS API, OpenAPI description, server-side auth, and documentation | A fresh client can complete the end-to-end sandbox flow |
A coding agent such as Claude Code or Codex can scaffold this quickly, but the agent cannot supply commercial rights, accurate club inventory, customer support, or approval. Those are business-system responsibilities.
Download the Muse Connector Prototype Prompt
The episode's starter brief is preserved as a local plain-text download so you can use it with your preferred coding agent.
Download the padel connector prototype prompt
The original companion resource is also available at Greg Isenberg's Muse Connector Prompt link.
Prepare for Behavior Review, Not Just a Form
Meta says it performs end-to-end testing. That makes awkward requests part of the product, not cleanup work for later.
- Authentication: reject missing, expired, and wrong-user credentials without leaking data.
- Authorization: enforce access on the server for every read and write operation.
- Unavailable inventory: return alternatives or a clear failure instead of inventing availability.
- Quote expiry: prevent a stale price from becoming a booking.
- Idempotency: make repeated reservation requests safe.
- Human approval: place the approval immediately before the consequential action.
- Recovery: document cancellation, modification, refund, support, and partial-failure behavior.
- Observability: preserve request IDs, quote IDs, reservation IDs, timestamps, and audit records.
- Public readiness: provide product, documentation, privacy, terms, support, access limits, and realistic example prompts.
Stripe says Muse asks users to approve the transaction total in chat and does not expose their underlying payment details to Muse. That payment protection does not remove the connector's responsibility to state the product, total, currency, cancellation terms, and merchant accurately.
Choose an Idea With Evidence
| Question | Good early signal | Warning sign |
|---|---|---|
| Can you reach the customer? | You can name ten buyers or one credible partner | The plan begins and ends with directory traffic |
| Is the task frequent or urgent? | Weekly repetition or a costly time-sensitive event | A novelty used once with no reason to return |
| Is supply controllable? | Authorized source with live state and support | Scraped or stale availability you cannot correct |
| Can the outcome be verified? | Stable lead, quote, booking, or shopping-list state | A vague recommendation with no completion record |
| Does the unit economics work? | Value per completed task covers fulfillment and support | Model and API cost look low while human exceptions are ignored |
| Can the connector survive platform change? | The same service works through web, API, MCP, or another agent | All demand, identity, and data live inside one platform |
A Seven-Day Starting Plan
- Day 1: interview three customers about the last time they completed the task, including delays, cost, and exceptions.
- Day 2: write one user request, one successful result, and five failure cases.
- Day 3: assemble a manual sample with clearly sourced or fictional data and ask for a buying decision.
- Day 4: define API objects for search, quote, approval, action, and status.
- Day 5: build the sandbox flow and test expired quotes, duplicate requests, and wrong-user access.
- Day 6: let a fresh reviewer follow the documentation without help and record every failure.
- Day 7: decide whether to pursue supplier access, revise the offer, or stop before investing in public submission.
The useful first milestone is not approval. It is one customer task that can be completed reliably, profitably, and safely through a documented interface.
Video Chapters
| Time | Chapter | Time | Chapter |
|---|---|---|---|
| 00:00 | Introduction | 01:35 | The App Store parallel |
| 04:55 | How a Muse connector works | 07:29 | Lead generation for suppliers |
| 09:28 | Home-repair dispatch | 11:07 | Padel match and court finder |
| 12:47 | Family dinner planning | 14:10 | How to choose an idea |
| 14:56 | How people find a connector | 15:32 | Partner with creators |
| 16:17 | Product-led sharing | 17:46 | Marketplaces and directory placement |
| 19:34 | Building the first version | 21:38 | Custom connectors and Meta approval |
| 23:56 | Where to start this week | 25:31 | Closing thoughts |
Sources and Useful Links
- Greg Isenberg: Meta Muse AI Connectors - The Next App Store Moment?
- Startup Ideas Podcast: Muse Connector Prompt
- Local download: padel connector prototype prompt
- Meta: Muse Connector Platform
- Meta: Muse product overview
- Stripe: Link payments and purchase approval in Muse
- Duffel: search, booking, and travel management through Muse
- Ticketmaster: live-event discovery through Muse
- Instacart Developer Platform
- Stacktree: firsthand Muse submission-form walkthrough
- Apple: the App Store's first decade
The video was published on 24 September 2026. Meta's platform page and supporting sources were checked on 25 September 2026. Connector requirements, review timing, availability, ranking, fees, and technical interfaces can change. Verify the live Meta form and current supplier terms before building or submitting.