How to Submit an MCP Connector to Meta Muse: The Complete Checklist
To list a product in Meta Muse, you submit a connector at muse.ai/platform: either a hosted MCP endpoint or a raw API, plus branding, policies and example prompts. Meta then tests it end to end for functional, security and legal requirements. Here is everything the form asks for and how to get it right the first time.
Short answer: open muse.ai/platform, click Submit a connector, sign in with a work email, and complete three steps: Overview, Technical specs and Review. You need a live, publicly reachable endpoint. Meta tests connectors against the real service rather than an uploaded package.
What is a Muse connector?
A connector links Muse to an app or service so the agent gains an ability, such as reading a calendar, sending an email or checking a website. Without one, Muse falls back to operating your website in a browser, which is slower, breaks more often, and some sites actively block it. With one, your service becomes a tool the agent can call directly.
Meta opened the platform on September 18, 2026, ten days after Muse launched. Its platform page describes the review as covering "functional, security, and legal requirements" with "end to end testing." Approved connectors appear in Muse's directory and can be featured editorially.
Step 1: Overview
| Field | What to enter |
|---|---|
| Connector name | Your product name as users would say it to an agent. |
| Company or developer | Legal or trading name, matching your privacy policy. |
| Product website | Canonical homepage URL. |
| Example prompts | Real requests, one per line, that map to your tools. For example: "Check whether my site is agent-ready." |
| Connector icon | 512×512 PNG or SVG. Use a full-bleed square with safe padding, because directories apply their own masks. |
| Payments | Whether the connector takes payments. Meta has partnered with Stripe so connectors can accept payments through Link. |
| Your name and work email | The reviewer's point of contact. |
| Support email or URL | A monitored channel. Reviewers and users will use it. |
| Privacy policy and terms | Live URLs. They should mention AI agent access and what data your tools receive. |
Step 2: Technical specs
You choose a connection type:
- Raw API: the API base URL, an optional OpenAPI spec URL, and documentation.
- Existing MCP: your hosted MCP endpoint URL and documentation. This is the faster path if you already run an MCP server, because the tool definitions are already machine-readable.
You also describe access requirements and select authentication methods. Developers who have filed, including Stacktree's write-up, report checkboxes for API keys, OAuth with PKCE, and Other.
Pre-flight checklist for an MCP endpoint
Reviewers test the live service, so fix these before you submit:
- Use HTTPS, publicly reachable, with no IP allowlist. Hit the endpoint from an outside network.
- Handle malformed requests correctly. Bad JSON should return JSON-RPC error
-32700, not an HTML error page. - Make discovery paths honest. If you don't serve OAuth metadata,
/.well-known/oauth-authorization-servermust return a real 404. Single-page-app hosts often return an HTML 200 for unknown paths, which breaks MCP clients. - Write tool descriptions for a model, not a human. Say what each tool does, when to use it, and what it returns.
- Return honest results. If a check can't be completed, say so. Never return a fabricated success.
- Test with the official MCP Inspector before submitting.
- Publish human-readable docs at a stable URL. That is what goes in the documentation field.
Step 3: Review
The last step reads back every field for editing and asks for three confirmations before submission. After that, you wait. Meta has not published a review time. Developers who filed early were told to expect weeks, not days, and Meta said on September 23 that it had received more than 1,500 applications in less than a week.
Common reasons a connector stalls
- The endpoint times out or needs a VPN.
- OAuth redirect URIs don't accept the platform's callback.
- Example prompts don't match any tool, so the reviewer can't exercise the connector.
- The privacy policy doesn't cover agent access or data retention.
A worked example
The All Clear Digital AI Agent Readiness connector is a single MCP endpoint. It checks whether AI assistants are allowed in by robots.txt, whether /llms.txt is a real text file, whether the homepage carries Organization or LocalBusiness schema, and whether FAQ content is present. Each check returns PASS, FAIL or UNVERIFIED rather than guessing. Its docs page shows the level of detail reviewers need.
New to Muse? Start with What Is Meta Muse? If you'd rather be found by Muse without building a connector, read how to make your website visible to Meta Muse.
Frequently asked questions
How do I submit a connector to Meta Muse?
Go to muse.ai/platform, click Submit a connector and sign in with a work email. The form has three steps (Overview, Technical specs and Review) and asks for a connector name, website, example prompts, a 512×512 icon, support and legal URLs, and either a hosted MCP endpoint or an API URL.
Does Meta Muse support MCP servers?
Yes. The technical step lets you choose between a raw API (with an optional OpenAPI spec) and an existing MCP server, in which case you supply the hosted MCP endpoint URL and its documentation.
What authentication does Muse accept for connectors?
Developers who have filed report checkboxes for API keys, OAuth with PKCE and Other. If your server requires user sign-in, OAuth with PKCE is the standard choice for MCP.
What size should the Muse connector icon be?
The form asks for a 512×512 PNG or SVG. Use a full-bleed square with the mark well inside the edges, because directories often apply their own rounded or circular mask.
How long does Muse connector review take?
Meta has not published a timeline. Developers who submitted in the first week reported being told to expect weeks, not days. Meta received more than 1,500 applications in under a week.