Meta Muse connectors: how to bring your business into the conversation

We spent years learning how to get our businesses in front of people on Facebook and Instagram. We built pages, published content, ran campaigns and answered messages. As customers changed how they discovered products, businesses had to change how they reached them.

I think we're at the beginning of another change like that. This time, the customer may ask a personal AI agent to find the right product and help them buy it. The business needs to be somewhere that agent can actually use.

That's why Meta Muse connectors matter to me. Meta launched Muse on 8 September as a personal AI agent and has opened a connector submission route for businesses. A connector gives Muse a defined way to work with another product. Meta's launch announcement and the Muse Connector Platform describe the product and review process.

At Noodle Seed, we already help businesses build for ChatGPT and Claude. We'll support Muse as well. The opportunity is to make your product useful wherever a customer asks their assistant for help.

Look at the audience Meta already reaches

Meta reported 3.60 billion people using at least one of its family of apps each day in June 2026. That includes Facebook, Instagram, Messenger and WhatsApp. It gives us a sense of the scale of the company bringing Muse to market. Meta's Q2 2026 results

Look at the individual products too. WhatsApp says more than 3 billion people use its service. Instagram reached 3 billion monthly active users, confirmed in Meta's Q3 2025 earnings call. Facebook reported 3.07 billion monthly active users in December 2023, a historical app-level figure rather than a September 2026 count. WhatsApp, Instagram disclosure, Facebook disclosure

Those audiences overlap. We cannot add them together, and they aren't Muse users. But they do show how widely people have adopted services that once asked them to learn an entirely new behaviour.

My view is that, over time, a majority of the people who made social and messaging apps part of everyday life will also use a personal AI agent. Muse could be one of those agents, alongside the evolving experiences from ChatGPT, Claude and Gemini. That's my expectation about the direction of adoption, not a measured conversion rate or a claim that most Meta users will choose Muse.

People won't need to understand the technology to make that change. They'll need it to save them time: find a useful product, compare a few options, check a delivery date, book an appointment or handle a repeat order. Those are familiar needs. What's changing is who does the work between the request and the result.

Social built the habit. AI is building the next one.

The interesting comparison is how quickly another everyday habit is forming. Google's Gemini app went from more than 400 million monthly active users in May 2025 to more than 1 billion in August 2026. OpenAI disclosed more than 300 million weekly active users in January 2025 and more than 900 million in February 2026. That's substantial adoption before every customer is delegating everyday purchases. Google's May announcement, Gemini's billion-user milestone, OpenAI in January 2025, OpenAI in February 2026

Two waves of adoption

Social reached billions.
AI is reaching them too.

Reported audience milestones · 22 September 2026

Earlier milestoneLater milestoneAI application

Monthly active users

01B2B3B

Facebook

Monthly active users · historical disclosures

Dec 20222.96BDec 20233.07B

Instagram

Monthly active users

Oct 20222B+Q3 20253B

Snapchat

Monthly active users

Feb 2023750M+Q2 2026971M

Gemini

Monthly active users

May 2025400M+Aug 20261B+

A weekly habit, too.

ChatGPT reports weekly activity. Shown separately, on its own scale.

0500M1B

ChatGPT

Weekly active users

Jan 2025300M+Feb 2026900M+

The next chapter

Now, the agent acts on your behalf.

For a business, the opportunity is to be part of what the customer can ask their agent to do.

Meta MuseLaunched 8 September 2026. No active-user total in the sources reviewed. Meta’s existing reach is a potential distribution advantage.Launch announcement

ClaudeAnthropic reported a sharp rise in consumer usage in April 2026, without a comparable active-user total in that disclosure.Company disclosure

These are adoption milestones across different periods, not a race on a shared timeline. Weekly and monthly audiences aren't interchangeable, and using an AI app doesn't necessarily mean authorising an agent to buy something. What interests me is that businesses can already see a large audience forming around a new way to ask for help.

From getting attention to being useful at the moment of intent

On a social feed, a business often starts by earning someone's attention. A customer might see a product, visit the profile, open a website and decide what to do. With a personal agent, the starting point can be a request: “Find a carry-on suitcase under $150 that will arrive before Friday.”

The agent needs more than a good description. It needs the current price, the right size, stock, delivery information and a supported way to help the customer purchase. If your business can provide those things, it can become useful at the moment someone is already trying to act.

Imagine the same shift for a local business. Someone opens Muse and says, “I need an appointment after work on Thursday.” They haven't picked a business yet. If you have the right appointment, can their assistant find it, explain the price and help them book it?

That is the distribution opportunity I would prepare for: millions of people asking their own AI agent to do things on their behalf, and businesses making their products available to those requests. It isn't a promise that one connector reaches millions of buyers. Each platform still controls access, discovery and review, and customers still choose which services to connect and what to approve.

If you want your business to participate in that channel, make it possible for an authorised agent to understand what you offer and take the next step with your product. A connector is one practical route into that experience. Search visibility can help someone find you; an integration can help their assistant use you.

Your product can become part of the answer

A website gives someone a place to learn about your business. A connector can give the assistant a way to use it. The customer can move from a question to a relevant option, and then towards a next step your business actually supports.

Think about a hotel. “We have rooms” is a starting point. “We have a room for your dates, this is the total price, and these are the cancellation terms” gives someone enough information to make a decision. If the booking system supports it, the assistant can then help prepare a reservation for the customer to approve.

The same idea applies to a software product. An existing customer might want to find a project, check its status or update a record. Giving them a way to do that through the assistant makes your product easier to return to.

These are examples of what a business could build. They aren't claims that these journeys are already available through Muse.

From a question to a next step

“Find me an appointment after work.”

An illustrative customer journey through Muse.

1. 01Find

   Your business shares available times and current prices.

2. 02Review

   The customer sees the appointment, price and terms.

3. 03Confirm

   Your system checks permission and attempts the approved booking.

A proposed appointment becomes a booking only when the business system confirms it.

There are three opportunities I'd look at

Help the right customer find you. A person describes what they need. Your product needs to offer information that helps the assistant judge whether it fits. Current availability, clear requirements and useful descriptions all matter.

Help them take the next step. Once there's a suitable option, remove the unnecessary work between choosing and acting. That might mean preparing a booking, starting a supported checkout or taking the customer to the right place in your product.

Make coming back easier. A customer who already uses your service may want to check an order, manage a booking or repeat a task. A connector can make that ongoing relationship more convenient.

I would treat each of these as something to test. A directory listing doesn't guarantee customers, and a conversation doesn't prove revenue. The opportunity is real enough to explore; the results still need to come from your own business.

Start with one customer request

The first decision is what you want someone to accomplish. “Connect our entire product” is a large project. “Help a customer find an appointment that fits their schedule” gives the team something specific to build and evaluate.

I'd start with a task customers already ask about. Where do they need help choosing? Where do they leave because a small question goes unanswered? Where does your team repeatedly look up information or explain the same next step?

Then write a brief your business and engineering teams can both understand:

• The request: “Find an appointment after 6 pm.”
• The information: available times, location, current price and cancellation terms.
• The next step: prepare an appointment for review, then attempt the booking after approval.
• The business result: a booking confirmed in your own system.

That last point matters. If your system can only request an appointment, the assistant should describe it as a request. If no times are available, it should say so. A good connector makes your actual service easier to use.

Where Noodle Seed fits

Your business already has products, prices, availability and rules. Noodle Seed helps your team turn selected things your product can do into tools an AI assistant can use. That might be finding an item, checking available appointments or preparing a purchase for the customer to approve. Noodle Seed documentation

You bring the product, the business rules and access to the systems behind it. Your team uses Noodle Seed to build, test and host that connection. The shared foundation can then support work across different AI applications, with setup and testing for each one.

Noodle Seed already enables building for ChatGPT and Claude. Muse support is planned. Preparing the product's capabilities now is useful work; approval in Muse's directory remains Meta's decision.

A new route to your customer

From “look at this” to “help me do this.”

Social discovery

Facebook · Instagram · Snapchat

1. See a post
2. Visit the business
3. Choose and buy

Agent-assisted action

Muse · ChatGPT · Claude · Gemini

1. Describe a need
2. Agent checks suitable products
3. Customer approves the next step

Make what your business offers available to an assistant: help a customer choose, book and buy.

ChatGPT and Claude build paths available. Muse support planned.

What does getting into Muse involve?

Meta's connector platform asks businesses to describe their product and submit it for review. Meta says it checks functional, security and legal requirements and completes end-to-end testing. Approved connectors can appear in the directory; featured placement is a separate editorial decision. Muse's submission process

Before your team starts, gather a few practical things: the customer task, examples of how someone would ask for it, the product website, support details, and privacy and terms pages. Your technical team will also need a working connection to the underlying product and a way to test the experience.

Be clear about who can use it. Does someone need an account, a paid plan or access in a particular region? What can the assistant read, and what can it change? Answering those questions early helps the team design a useful experience and explain it honestly to customers.

If you want to see how that becomes an MCP server, switch to the Technical view at the top. It walks through the tools, customer access, testing and submission preparation.

Measure the booking, not just the conversation

For our appointment example, I'd want to know how many customers found a suitable time, how many reviewed a proposal and how many ended up with a confirmed booking. I'd also want to see failed attempts and whether customers came back.

Ten availability searches and three confirmed bookings tell a different story from thirteen successful API requests. Your booking system tells you which appointments actually exist. That's the evidence I'd use to judge whether the connector is helping.

Where Muse provides discovery or referral data, use it. Where it doesn't, leave that part of the journey unknown. You can still learn whether the customer task works without pretending to know exactly how every customer found you.

Bring one useful thing your product can do

Businesses learned to show up where their customers spent time on social media. I think the next step is to show up where their customers ask an agent to act. Start with one thing someone wants to accomplish, make the information reliable, and give them a next step they can approve.

That is how we're thinking about Noodle Seed: helping businesses grow conversationally, with an assistant that helps customers choose, book and buy. ChatGPT and Claude are already part of that work. Muse will be too.

Explore Noodle Seed with one customer journey in mind. If your team is ready to build, the developer quickstart is the starting point.

Suppose your customer asks Muse to find an appointment after work. Your API already knows which times are available. The engineering task is to let the assistant use that information, prepare a clear proposal and make a booking only after the right checks and approval.

That's the connector I'd start with. One customer task, a few well-defined tools and a result your business can verify.

Meta's Muse Connector Platform offers a reviewed route into its directory. Noodle Seed already enables building for ChatGPT and Claude, and Muse support is planned. This walkthrough explains the MCP foundation you can build now and the Muse-specific work that still needs verification. Muse Connector Platform

Where MCP fits into Muse connectors

MCP stands for Model Context Protocol. It gives an AI application a common way to discover and call capabilities exposed by another system. Your API is how software talks to your business. An MCP server makes selected capabilities available in a form an AI assistant can understand and use. MCP's server concepts

For example, your API may have separate endpoints for products, variants, inventory and delivery. A customer asks, “Can you find me a waterproof jacket in medium that arrives before Friday?” The connector needs to help turn that request into the relevant checks and return enough detail for a useful decision.

There is evidence of an MCP submission route for Muse. In its firsthand account of submitting a connector on 19 September, Stacktree reports that Meta's form offers “Existing MCP” and “Raw API” connection types. It describes fields for a hosted endpoint, documentation, access requirements and authentication. We've inspected Meta's signed-in Overview form; the technical-field details here come from Stacktree's submission account, rather than our own completed submission.

This is encouraging if you've already built an MCP server. It gives you something concrete to prepare for review. It doesn't establish that every MCP feature, authentication flow or embedded interface will work identically in Muse, ChatGPT and Claude. Each host still needs testing.

There is also a separate route: Meta says Muse can create custom connectors for a person using services with APIs or command-line tools. That's different from getting your business reviewed and listed in the directory. For this article, we're focusing on the business distribution route. Meta's technical explanation

What actually happens when an agent uses your product?

Suppose we're designing a connector for an appointment business. A customer asks, “Find me an appointment after work on Thursday.” A useful implementation would work through the following steps:

1. Understand the request. Get the service, location, date and time zone. Ask for any missing detail that changes the search.
2. Check the business system. Read real availability through the API and return a small set of suitable options.
3. Prepare the choice. Show the selected service, time, price and relevant cancellation terms.
4. Get approval. Let the customer review the exact appointment being proposed.
5. Try the booking. Recheck availability and permissions, then ask the booking system to reserve it.
6. Report what happened. Show a booking reference only when the system confirms one. If the result is uncertain, check its status before attempting another booking.

This is an illustrative design, not a description of a shipped Muse booking integration. It shows why the connector needs more than a convincing answer. It needs access to current information and a reliable way to handle the action.

Your business keeps responsibility for its rules. If the customer doesn't have permission to change an appointment, the API must reject the change. If the price changes, the earlier approval should no longer cover the new proposal. Meta's Sentinel controls permission for actions leaving Muse, but your service still needs its own checks. How Meta built safety into Muse

How to build the foundation with Noodle Seed

Noodle Seed gives you a way to build, validate and host the MCP server that exposes your product's capabilities. Our current developer workflow supports ChatGPT and Claude. Muse is the next surface we plan to support; there isn't a verified Noodle Seed-to-Muse deployment walkthrough in this article. Noodle Seed developer documentation

The useful work can start now.

1. Choose one customer task

Write down one sentence a customer might actually say. “Find an appointment after 6 pm” is a good starting point. Then decide what information the agent needs and what the customer should be able to do next.

Keep the first version small enough to test properly. You can learn a lot from a connector that checks availability and prepares a clear next step before giving it permission to create bookings.

2. Connect it to the real source of information

Identify the API that owns availability, prices and booking records. Gather its documentation, a test account and the permission scopes the journey requires. Confirm which actions the API supports before designing the conversation around them.

If the API can only request an appointment, call the result a request. If it returns no availability, distinguish that from a failed lookup. Those details determine whether the customer can trust the experience.

In Noodle Seed's technical vocabulary, a connector can also mean the API connection behind your MCP tools. The Muse connector is the integration a person enables in Muse. These are different parts of the same path. Noodle Seed's core concepts

3. Build tools around the task

Give each tool a clear purpose, defined inputs and a small, useful result. Keep looking up information separate from changing something. Noodle Seed's tool-design guide explains this approach.

For our appointment example, I'd start with these operations:

These names describe a proposed business contract, not built-in Noodle Seed tools. Your implementation must connect them to your system and enforce their behavior.

A result should make the state easy to understand. For example:

{
  "proposal_id": "example_proposal_42",
  "status": "awaiting_confirmation",
  "service": "Consultation",
  "starts_at": "2026-10-01T18:30:00+01:00",
  "time_zone": "Europe/London",
  "price": { "amount_minor": 4500, "currency": "GBP" },
  "cancellation_policy": "Free cancellation until 24 hours before",
  "expires_at": "2026-09-21T19:10:00Z"
}

This is synthetic example data. It is a proposal, so it deliberately has no confirmed booking reference. The server should store the proposal and bind it to the authenticated customer. Confirmation should resolve that stored record, recheck its validity and use a stable request key to prevent retries from creating duplicates. A field saying confirmed: true in model-generated input isn't proof of customer approval.

4. Use Noodle Seed to build and test it locally

Start with Node.js 24 or newer and the Noodle Seed quickstart. For a new project, the documented bootstrap is:

npx --yes @noodleseed/one@latest init my-muse-connector
cd my-muse-connector

The name is just your project name; it doesn't enable a special Muse mode. Follow the generated project instructions and use the Noodle Seed developer plugin in your coding agent. A useful first brief would be:

Build a Noodle Seed MCP server for our appointment business using the API documentation I provide. Start with finding available slots and preparing a booking proposal. Keep booking confirmation separate and disabled until we verify customer authorization and approval. Return current prices, time zones and cancellation terms. Use test data only where the backend is unavailable and label it clearly. Keep credentials out of source code. Test expired access, unavailable slots and uncertain booking outcomes. Prepare the server for future Muse review without claiming it is already supported or listed.

After implementation, run the project's generated checks and local development workflow. Test the actual customer questions as well as the tool inputs. Synthetic tests help check your code; a sandbox or real test account is still needed to prove that authentication and the backing API behave as expected.

5. Give each customer the right access

The connector must know whose account it is using. Signing in proves identity; authorization decides which records and actions that identity can access.

Noodle Seed verifies configured customer tokens and can restrict tools by roles and scopes. With direct or federated OIDC, your application still owns the authorization server that signs people in. Noodle Seed doesn't automatically supply that missing system. Customer authentication in Noodle Seed

Use the authenticated identity to resolve the business account. Don't let a tool argument choose an arbitrary customer's account. Keep access limited to the task, and test what happens when someone disconnects or loses permission. For a booking, test the approval flow in the actual host before enabling writes there.

6. Deploy a tested endpoint, then prepare the Muse submission

Once the workflow is ready, use Noodle Seed's deployment process to obtain a hosted MCP endpoint. Confirm the intended access configuration: the documented default is owner-only, which is not a customer launch. Noodle Seed deployment quickstart

For Muse, start at the official connector platform. Its Overview form asks for your connector name, company, product website, example prompts, a 512 × 512 PNG or SVG icon, payment category, contact and support details, and privacy and terms URLs. Prepare those alongside technical documentation and a testable endpoint. Stacktree's account describes authentication and access-restriction fields in the next stage; check the current requirements when you submit. Firsthand technical-form walkthrough

Explain any account, subscription or regional requirements before a reviewer or customer hits them. Test discovery, sign-in, tool use, approval, error recovery and disconnection in Muse when access allows. Keep embedded UI and payments as separate compatibility checks. An MCP endpoint doesn't establish either one.

Meta decides whether the connector is accepted. Noodle Seed provides the foundation for building and operating your product's capabilities; it cannot promise directory approval or placement.

One product foundation

Build the capability.
Verify every destination.

01 / Your businessThe API

Products, availability, prices and business rules.

02 / Noodle SeedYour MCP server

Useful tools, customer access and a hosted endpoint.

03 / AI applications

ChatGPTSupported build path

ClaudeSupported build path

Meta MusePlanned

Each destination needs setup and testing. Meta reviews Muse directory submissions.

Build once, then verify each place your customers use it

This is where I think MCP becomes valuable for a business over time. The work of deciding what a customer can do, connecting the right data and enforcing the rules should have a reusable home. You shouldn't need to rethink your booking logic every time another assistant becomes relevant.

That is the direction we're taking with Noodle Seed. We already enable businesses to build for ChatGPT and Claude, and we'll support Muse as well. The shared foundation can carry the product's capabilities, while each destination still needs its own setup, testing and, where applicable, review.

“Build once, deploy anywhere” is useful as a direction because it reduces repeated work. The responsibility is in making sure the customer journey works wherever you deploy it.

Measure what the customer actually accomplished

When the connector is available, I'd follow the journey from a useful request to a confirmed business result. Where platform data is available, compare discovery and connection with successful task use. Inside your own system, measure proposals, completed bookings or purchases, failed attempts and repeat use.

Use stable proposal and transaction references to avoid counting retries as new business. Keep personal data out of analytics unless it's necessary and permitted. If Muse doesn't expose impressions or a reliable referral signal, mark that part of the journey as unknown. Don't manufacture an attribution rate from tool calls.

For our appointment business, ten availability searches and three confirmed bookings tell a different story from thirteen successful API requests. The booking system is the authority on which appointments exist. That's the level at which I'd want to judge whether the connector is helping the business grow.

Questions businesses are asking

Does Meta Muse support MCP connectors?

Meta has opened a reviewed connector submission route. A developer's firsthand submission account reports an “Existing MCP” option alongside “Raw API.” That is evidence of a submission path, not proof that every MCP server or feature already works in Muse. Source

Can I connect an existing API instead?

The same submission account reports a Raw API route. An MCP server can still be useful when you want to expose a deliberate set of customer tasks and reuse that foundation across supported AI applications. Check Muse's current requirements before choosing the submission route.

Is Noodle Seed's Muse integration live?

Muse support is planned. Noodle Seed already enables building for ChatGPT and Claude. You can start preparing your product's MCP capabilities now; Muse compatibility and any directory approval still need to be verified separately.

Will a connector automatically bring customers?

No. It gives your product a way to participate, but growth depends on discovery, usefulness, reliability and what customers actually complete. Treat distribution as something to measure after launch.

Does a Muse connector automatically enable payments?

Meta's platform page references Stripe Link for payments. That doesn't establish payment readiness for your business. Confirm the supported payment flow, commercial requirements and transaction behavior separately before offering purchases. Muse Connector Platform