You Bring the API: Meta Opens Muse's Connector Platform to Developers

Ten days. That is how long it took Meta to go from launching Muse, its personal AI agent, to opening the product’s most strategically important surface — its connector platform — to third-party developers.

At 7:03 PM Eastern on September 18, 2026, Mark Zuckerberg posted: “Opening access for developers to build Muse connectors.” The post linked to muse.ai/platform, a one-screen page with a single button — Submit a connector — and a three-sentence pitch that is best read as a mission statement for the entire consumer-agent era:

“You bring the API — Muse brings the agent, the browser, and the context of what the person actually wants. People reach your service just by asking for it, and their agent takes it from there.”

Within two hours, the post had passed 230,000 views. By the next morning, the story was being picked apart across the developer community — less for what Meta published than for what it left out.

What actually launched

The platform page describes a process, not a program. Three numbered steps:

1. Describe your product — tell Meta what your connector does and how users will use it.
2. Submit for review — Meta evaluates “functional, security, and legal requirements” and runs end-to-end testing.
3. Appear in the directory — once approved, users find your connector inside Muse, where “editors will review connectors for featured placement.”

That last clause is doing quiet but heavy lifting: an editorial layer means featured placement inside a consumer agent — a new form of shelf space, ranked by Meta’s editors rather than by an algorithm users can see.

For payments, Meta says it has “partnered with Stripe to make accepting payments easy with Link” — the same payments rail Muse already uses for its own agent-driven checkouts, accepted at more than a million businesses.

What the page does not publish is the part developers actually need before committing engineering time:

• No SDK or API specification
• No fee or revenue-share structure
• No review timeline
• No developer terms of service
• No list of launch partners or example connectors

The submission form asks for a description of the product — not code — and the page does not say who writes the connector once the description is in.

Two kinds of connector, one word

Meta’s help center already describes two distinct things under the single word “connector,” and understanding the difference is essential to reading this launch correctly.

Directory connectors are what the new platform page opens up: submitted by a business, reviewed by Meta for functional, security, and legal compliance, listed under Muse’s Settings, and connected by the user with a tap.

Custom connectors have existed since launch: any user can ask Muse to build a connector to any service that has an API, and Meta builds it on the spot for that one user. Meta is unusually direct about the risk here — it “doesn’t review custom connectors or how they use your information, so grant access with caution and review the provider’s privacy policies.”

For businesses, this creates a peculiar dynamic: your customers can already reach you through Muse, terms or no terms, via the unreviewed custom-connector path. The reviewed directory is the way to meet them on Meta’s approved terms instead.

Why the timing is the real story

The sequencing this week was deliberate: audience first, then the shelf.

• September 17 — Muse for Mac launches, extending the agent from mobile and web onto the desktop, where it can interact with files, messages, calendar, notes, and mail in their native applications.
• September 18 (morning) — Business Insider and 9to5Mac report Muse at #1 among free apps on Apple’s US App Store, ahead of ChatGPT, with a 4.9-star rating across more than 13,000 reviews.
• September 18 (evening) — the developer door opens.

Ten days from consumer launch to a developer-platform pitch is fast by any platform’s history — the App Store took a year to add third-party apps, and Facebook Platform took three years from Facebook’s own launch. Meta is compressing that cycle because it can: the #1 chart position gives developers a concrete, current answer to “is anyone actually going to use this?” instead of a hypothetical future user base.

The connector list at launch, per Alexandr Wang’s September 8 thread, already reads like a normal person’s phone: Gmail, Google Calendar, Outlook, Google Docs, Plaid, OpenTable, Spotify, Function Health, Withings, Tailscale, Peloton — plus Muse-only connectors for Instagram (including DMs), Threads, Facebook (including Marketplace), and Messenger. The connectors Zuckerberg says went “live today” on September 18 are, notably, unnamed in every primary source.

The inversion: who owns the moment of choice?

Every app store since 2008 has worked the same way: the developer builds the client, the platform owns the shelf, the user finds and installs. Zuckerberg’s three sentences take the client out of the picture entirely.

Under the Muse model, the business keeps its service and its API. The agent — running on Meta’s computer, with Meta’s browser and Meta’s memory of “what the person actually wants” — comes to the service. Users reach a business “just by asking for it.”

Whoever owns the agent owns the moment of choice. That is the sentence every business evaluating this platform should read twice. It also explains the editorial featured-placement layer: in a world where the agent is the interface, being discoverable by the agent is the new SEO, and Meta controls that ranking.

The honest caveats

It is worth being skeptical of how simple the pitch sounds, because “you bring the API” glosses over real work. An API designed for a traditional UI — specific request sequences, pagination, form-based inputs — does not automatically become agent-friendly just because it is reachable. A good connector still requires designing how an agent discovers capabilities, handles ambiguity, manages side effects that need user confirmation, and surfaces errors conversationally rather than as raw HTTP codes.

There are also structural questions Meta has not answered:

• Security vetting: what the review actually checks, and what stops a connector from being built with poorly-scoped or malicious access, is not detailed.
• Single-vendor lock-in: unlike MCP, the open cross-agent protocol, a Muse connector reaches only Muse’s user base. If Meta changes terms, review policies, or revenue share later, developers have no fallback but to rebuild. An MCP server, by contrast, works across every compatible harness.
• Unresolved access-control questions: Muse’s Sentinel security architecture already faced scrutiny over what it guarantees on desktop; a third-party connector ecosystem adds another layer on top.

The pragmatic read for builders: a Muse connector is an addition to an MCP-based integration strategy, not a replacement for one — worth having for the reach, but not worth betting the roadmap on before the first wave of documented connectors shows how the review process behaves in practice.

What to watch

The signal that matters over the coming weeks is not Meta’s messaging but the shape of the first real connector wave. If early connectors skew toward calendars, tasks, and notes, Muse is being positioned as a personal-assistant extension layer. If they skew toward commerce and transactions — bookings, purchases, payments — Meta is pushing the more ambitious “agent completes purchases on your behalf” vision. Either way, the directory and its editorial rankings are about to become some of the most contested shelf space in consumer software.

Meta has spent ten days proving Muse has an audience. The next ten will show whether developers believe the audience is worth building for on Meta’s terms — with the terms themselves still unpublished.