Custom Connectors and MCP, Explained for Non-Developers

Meta's Muse reaches into Gmail, Slack, QuickBooks, and dozens of other apps through ready-made connectors. But what about the app it doesn't cover: your industry-specific tool, the internal dashboard your team built, the service with an API but no Meta partnership? That's what custom connectors are for. And the technology making them practical has a name you'll keep seeing: MCP.

You don't need to be a developer to understand either one. Here's the plain-English version.

What an MCP server is, simply

MCP stands for Model Context Protocol. Forget the jargon. Every app speaks its own private language. MCP proposes a standard plug for AI-to-app connections: one standard way for an agent to ask an app “what can you do?” and then do it.

Think of it like USB. Before USB, every device needed its own cable and port. USB standardized the connection. MCP is trying to do the same thing for AI assistants and software.

The 3 ways Muse connects to a service

1. Built-in connectors. Meta builds these with the service provider: Gmail, Google Calendar, Spotify, Plaid, Notion, and the rest of the Muse connector directory. Switch them on under Settings → Connectors. Each one ships with skills Meta wrote and tested.

2. Custom connectors. For everything else. If a service has a documented API, ask Muse to build a connector for it. No Meta partnership needed. Your credentials go into Muse's Secure Credentials Store, never into the chat. The catch: Meta does not review custom connectors. Grant access carefully.

3. The browser fallback. Some sites have no API at all. For those, Muse can drive a real browser on its cloud computer the way a person would. It works on almost anything, which also makes it the most fragile option. Keep the browser as a last resort.

How to get a custom connector: the chat flow

1. Ask in plain words. Tell Muse what you want: “connect to my [service]” or “build a connector for [service].”
2. Answer its questions. Muse may need the service's API documentation or details about how you log in.
3. Put credentials in the secure prompt, never the chat. Use the dedicated secure credential prompt.
4. Let it build and test. Muse writes the connector code, runs a test call, and saves the working connector.
5. Start read-only. Add write actions after you've watched a few clean runs.

People who build these regularly keep destructive actions two-phase until the integration earns trust: the connector proposes, you confirm.

Developers should know there's a formal path too: Meta accepts MCP-based connector submissions for the official directory through muse.ai/platform, where they go through functional, security, and legal review.

Keeping your credentials safe

• Credentials live in the Secure Credentials Store, separate from your chats.
• Meta doesn't vet custom connectors. Read the service provider's privacy policy.
• Start narrow. Read-only first, then widen permissions only for reliable connectors.
• You can disconnect anytime in Settings → Connectors.

When a custom connector beats a built-in one

Reach for custom when your tool isn't on the built-in list, you need a narrow slice of a service, or you already have an MCP server. Stick with built-in when one exists. If the service has no API at all, the browser fallback is the only path.

Custom connectors let you shape Muse into infrastructure of your own. Once you know the three tiers and respect the credentials boundary, start read-only.