How to Add a Remote MCP Server to Raycast

How do I add a remote MCP server to Raycast?

Open Raycast, run the Install MCP Server command, and fill in the form: a name, Transport: HTTP, the server's URL, and OAuth Type: Dynamic. Click Install, approve in the browser, and the server is available by @-mention in AI Chat and Quick AI. There is no config file to edit. MCP is not Pro-gated — Raycast's MCP page carries no Pro badge, and its manual states that users on Mac v2 and Windows get free access to a default model (GPT-5.6 Luna) during the Beta, subject to standard usage limits. Pro raises those limits and unlocks the wider model list; it is not what turns MCP on.

Requirements

• Raycast on macOS. No Pro subscription is required to use MCP — users on Mac v2 and Windows get free access to a default model (GPT-5.6 Luna) during the Beta, subject to standard usage limits. Pro buys higher limits and the wider model list, and some frontier models sit behind the Advanced AI add-on

Adding the server

Run the Install MCP Server command and fill in:

• Name — this becomes the @-mention handle, so pick something you will want to type
• Transport — HTTP
• URL — the server's endpoint
• OAuth Type — Dynamic, when the server supports dynamic client registration. That is what makes the setup keyless; no client ID or secret is needed.

Click Install.

The name field deserves a moment's thought. It is not a label — it is the handle you will type in every conversation, and renaming it later means retraining your own muscle memory.

Authorization

Your browser opens with the server's consent screen. Sign in if prompted, review, and approve. You are returned to Raycast with the server installed.

Using it

@-mention the server's name in AI Chat or Quick AI to bring its tools into the conversation.

To review, reauthorize, or remove the server later, run the Manage MCP Servers command.

iOS

Remote HTTP servers are the transport Raycast supports on iOS — its MCP documentation notes that iOS is always HTTP — so a hosted server is the kind that can work there at all, unlike a local stdio server.

What Raycast's manual does not state is that installing on macOS makes the server appear on iOS automatically: its Install and Manage sections carry separate Desktop and iOS tabs, which points toward installing per platform. Treat iOS as "supported, install it there too" rather than "carries over on its own," and check the iOS tab in Raycast's own docs if you need certainty.

That is still a real difference from editor-based clients, where the server is bound to that editor's install. Here the connection points at a hosted endpoint, so the phone can reach the same server rather than needing its own copy of it.

What usually goes wrong

• The @-mention doesn't autocomplete. Run Manage MCP Servers and confirm the server installed. The handle matches the name you gave at install — if you named it something else, mention that exact name.
• AI features aren't available at all. Not a Pro problem — Raycast AI is free during the Beta on Mac v2 and Windows, subject to usage limits. Check that you are on a supported platform and build, and whether you have hit the usage limit for the period.
• The OAuth window never appears. Re-run the install, and check pop-up blockers and your OS-level default browser setting.
• The connection fails immediately. Check the URL against exactly what the server publishes, including the path and any trailing slash.

A worked example

Raycast is a launcher — its AI is always one keystroke away, which changes what a memory server is useful for. Instead of context that lives in one editor, it is context you can reach from anywhere on the machine.

Tempreon is a hosted MCP server built for exactly that shape. Install it with the name tempreon, Transport HTTP, OAuth Type Dynamic, and this URL:

https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

Then @tempreon in AI Chat or Quick AI brings your context into the conversation, on the Mac and on your phone. The account-holder walkthrough is in the Raycast setup doc.