Meta now takes MCP servers from outside developers on Muse. You submit an existing remote server through one form on the Muse Connector Platform, and once Meta approves it, the connector is listed in Muse (Meta's connector guidelines).

The form is short. What decides the outcome sits behind it: the guidelines ask you to classify every tool, document it, and give reviewers a test account that stays live for the whole review. This guide covers the form first, then what to have ready.

Why Muse, and why now

People used to search for a service. Now they ask an agent for it. Muse is Meta's personal AI agent: it books trips, sends emails and fills in forms for the user, and asks before anything sensitive.

It reached number one on the US App Store and is one of only twelve apps ever to pass 10 million US downloads in its first 40 days. A listed connector is how your product gets picked when someone asks, with no landing page, no ad and no search ranking in between. Most of that audience is in the US for now, and Meta already has more than 4,500 submissions to get through, so the sooner yours is in the queue, the better.

Before you start

Muse connects from Meta's cloud, so your endpoint has to be a public HTTPS URL. A server on localhost or behind a VPN won't connect.

You can try your server in Muse before submitting anything. Ask Muse in a chat to create a custom connector for your server URL, sign in, then run the example prompts you plan to put in the form. Meta does not review custom connectors, so this is a private test and not a listing.

Classify your tools

Meta's guidelines sort every tool into one of three classes, and Muse's permission prompts follow from the class. Reads and non-sensitive writes ask the user on first use. Sensitive writes, such as completing a payment, ask every time, and the user can never set them to always allow.

A tool that mixes reads and writes counts as a write, and a payment, send, share or publish action can't hide inside a tool described as a read.

Put the class of each tool in your tool documentation, because reviewers check it and Meta approves only the tools it has reviewed. If your OAuth setup allows it, also offer users a Read Only option.

Submit the form

Go to platform.muse.ai, choose Submit a connector and sign in with your work email.

Overview. Enter the connector name, your company, the product website and example prompts, written as a user would type them to Muse. Upload the icon (512×512, PNG or SVG), choose whether the connector accepts payments, and add your contact, support, privacy policy and terms of service details.

Technical specs. Switch the connection type to Existing MCP, then enter the hosted MCP endpoint, the URL of your API or MCP documentation, and any access requirements, such as a paid plan or regional limits.

Review and submit. Check your entries and send the form.

Prepare for review

After the form, the guidelines list five things Meta needs before testing starts: the connector overview, business verification and terms, a data processing questionnaire, integration credentials with tool documentation, and a test account. The credentials are an OAuth client or an API key, marked as test or production.

The test account matters most. Reviewers connect with it and test each tool, so keep it accessible throughout the review, load it with representative data and include sign-in instructions. Meta then reviews in three stages: risk assessment, tool review and end-to-end testing, which includes manual review.

Two requirements catch people out. A connector has to finish the tasks it starts and improve on using the browser, and Meta says "a wrapper around public pages or a list of links alone is not enough". Meta is also not approving new tools that move money, such as transfers.

After you submit

Track the status in the developer portal. Meta can't answer individual messages about status. It expects a few more weeks to reach a healthy pace of onboarding, and it will email you if it needs more information. The guidelines carry no date and can change, so reread them before you submit.

If your server isn't hosted yet, ship with Alpic: import your Git repository in the dashboard and each deployment gets a URL to use in the hosted endpoint field. Connect GitHub and every push redeploys. We recommend one environment per submitted version, named after the release (v1, for example), so the version under review stays put while you work on the next one.