Meta Muse AI Connectors: The App Store for AI?

Big business opportunity: “App Store for AI” via Meta Muse Connectors

• Meta introduced Muse, a personal AI agent for developers.
• Meta also allows developers to submit “connectors”—integrations that let Muse access your service when a user asks for help.
• The video frames this as an App Store moment:
  • Meta provides the assistant + distribution
  • External businesses provide the actions/services that get executed on the user’s behalf
• Market-formation hypothesis: if Muse becomes widely used for tasks requiring third-party services, “billions of dollars” could flow through these integrations (distribution + transaction execution).

---

How Muse connector workflows are supposed to work (end-to-end)

Example: podcast studio booking

1. User intent
   • “Get me a podcast studio on Tuesday afternoon for 2 hours, budget $200.”
2. Muse determines context
   • Location/city
   • Recording type requirements
3. Connector calls the business via an API (described as a “service window”)
   • Availability check for a specific date/time
   • Pricing/quote (e.g., available room at ~$160 including equipment)
4. User approval
   • Muse shows the offer/quote
   • After approval, the connector triggers booking
5. Business handles operational realities
   • Calendar accuracy / availability correctness
   • Equipment readiness / door access
6. Output back to Muse
   • Confirmation details

---

Key playbook: start with “the smallest complete customer task”

Founder approach recommended:

• Write down one thing the customer should accomplish, e.g.
  • “Show available courts near me tomorrow evening with prices.”
• Use a coding agent (e.g., Claude Code / Codex) to implement:
  1. inventory/availability check + pricing quote first
  2. then the booking operation once quote flow works
• Connector requirements:
  • Hosted somewhere Muse can reach
  • Secure customer identity/access checks so users can only access what they’re entitled to
  • Robust handling of edge/“awkward requests” to pass review

Review-stress examples (to handle in advance)

• Requesting already-booked times
• Using expired codes
• Repeated requests that should not create duplicate reservations
• Common follow-up changes (e.g., “move booking to another day”)

---

Business discovery + execution framework (recommended process)

1) Customer discovery interview

• Prioritize talking to customers.
• Identify a customer type you can reach.
• Ask about the last time the task happened.
• Learn:
  • how they got it done
  • where they had to wait
  • what it cost

2) MVP slice

• Implement only a small part of the job as a connector.
• Validate whether Muse can complete the task end-to-end.

---

Connector submission & approval: operational considerations

• Meta has:
  • a submission process
  • a directory of approved connectors
  • directory + editors who may feature connectors
• Uncertainty noted by the video:
  • how many connectors Meta approves (daily/weekly pace)
  • how often unknown connectors get recommended during general conversation vs. via directory exposure
• Integration submission paths mentioned:
  • public listing
  • custom connector option to connect an AI via your service’s API info first (try integration before public listing)
• Review risk guidance (build rejection resistance by testing):
  • already-booked time requests
  • expired codes
  • duplicate/invalid request handling
  • follow-up modifications to existing intents (e.g., rescheduling)

---

Marketing / distribution playbook (how people will find and use your connector)

Primary channel (uncertain but potentially strong)

• Meta connector directory (and editorial featuring)
  • Compared to App Store feature placement driving massive downloads
  • Example cited: ~40,000 downloads/day during a feature

Supplemental / safer channel: build your own demand

1. Work with existing creators/audiences

   • Example: a family dinner connector powered by a creator who posts recipes
   • Creator demonstrates a task (turn tested meals into a grocery order)
   • Agree on payment terms for introduced conversions

2. Product-led growth / viral sharing loops

   • Example: paddle/match booking where a player can share a reservation page with others
   • Framed loop:
     • multiplayer product → easier sharing → others convert → more bookings

3. Distribution through connected marketplaces

   • Example concept: Ticketmaster showing events via connectors
   • Recommendation: if your industry has marketplaces, study how businesses participate and what data helps customers decide.

4. Relying on directory feature placement (not dependable)

   • Don’t assume random editor/product-manager selection
   • Use other distribution levers too

---

Four concrete connector startup ideas (business models + how they would work)

1) Lead gen for business suppliers (B2B subscription)

Concept

• Example supplier: linen service for restaurants.
• Muse finds timely openings (e.g., restaurants opening nearby).
• Connector returns:
  • verified upcoming openings
  • sources showing when openings are expected
  • public contact when available

Differentiator

• Timeliness + verification vs. generic lead lists.

Operational need

• Keep data current (openings can delay or change).

Revenue model (example)

• Prospects choose lead cadence: weekly / every 2 weeks / every 3 weeks
• Hypothetical: 100 customers × $99/month → ~$9,900 MRR (before costs)

---

2) Home repair dispatch (lead/booking fee)

Concept

• User: “Dishwasher broken—Samsung model—someone can come tomorrow.”
• Connector checks participating repair providers for:
  • exact appliance/model match
  • availability tomorrow

User chooses provider

• Connector then:
  • passes job details, or
  • books the appointment where supported

Pricing (illustrative)

• Charge repair companies a fee like $100 for a qualified intro or confirmed booking.

Go-to-market guidance

• Start with one repair type + one city/town
• Build knowledge of what providers truly handle
• Track local demand cadence (providers need daily/regular bookings)

---

3) Local paddle match + court finder (marketplace bookings)

Concept

• User: “Find a paddle game tonight near me with people around my level.”
• Connector searches:
  • participating clubs/courts
  • available timeslots
• User reserves; if they already have a group, they can reserve court time and share details with friends.

Operational / growth assumptions

• Video expects Muse to become more social over 6–12 months, which may amplify multiplayer sharing.

Monetization

• Fee on bookings with partner clubs.

Adjacent product reference

• Playtomic cited as a platform for court/match booking; opportunity is filling gaps where availability requires manual checking.

---

4) Family dinner planning + grocery integration (subscription + retention via iteration)

Concept

• Parent request: plan dinners for next 3 nights, max 20 minutes cooking/day, already have rice + broccoli.
• Connector:
  • generates plans from tested recipes tailored to family constraints
  • adjusts quantities
  • outputs missing items as a shopping list

Grocery partner integration

• Instacart dev tooling can produce a shoppable list experience.

Testing + improvement loop

• Subscription families cook weekly.
• Learn:
  • what they actually cooked
  • what worked / was disliked
• Use feedback to improve future meal plans.

Potential exit speculation (high level)

• Instacart could acquire a connector if traction grows.

---

Metrics / KPIs mentioned or implied (and how to track them)

Explicit benchmarks/examples

• Apple App Store example
  • By June 2010, paid apps + in-app purchases paid developers $1B in <2 years
  • Used as a distribution economics benchmark
• Directory-feature impact example
  • ~40,000 downloads/day after being featured
• Lead-gen model example
  • 100 customers × $99/month → ~$9,900 MRR
• Travel connector scale benchmark
  • Duffel reported passed $1B annual transaction value

Implied operational metrics to run

• Connector usage frequency
  • how often users complete tasks via Muse
• Unit economics funnel
  • recommendation → request → approval → booking/transaction conversion
• Lead effectiveness
  • rate of “qualified leads” (not just volume)
• Retention
  • weekly/biweekly meal usage for family subscriptions
• Reliability
  • booking accuracy
  • handling duplicate/invalid requests (review gate + user trust)
• Distribution performance
  • creator partnership impact + sharing loops
  • downstream conversion rates

---

Presenters / sources mentioned

• Greg Eisenberg (host; Startup Ideas podcast)
• Apple (App Store historical example)
• Meta / Muse (product described)
• Duffel (travel example; Muse integration claim)
• Thumbtack (lead-gen market validation reference)
• Apollo (general contact search reference)
• Playtomic (paddle booking reference)
• Instacart (shoppable list integration reference)
• Ticketmaster (marketplace/distribution analogy)
• Claude Code and Codex (coding agents referenced)
• MCP (Model Context Protocol) (connector/integration standard referenced)
• Stack Trees founder (mentioned for publishing an account of Meta review form experience)
• ideabrowser.com (startup ideation resource mentioned)