MacServe.

MacServe · Early access

Use an AI assistant

Authorize selected platform features, then deploy through SSH.

Early-access connection

Remote MCP address: https://panel.macserve.io/mcp. Start authorization in your assistant, sign in to MacServe, choose the Macs and permissions, then return to the assistant. Only approve a request you started; client names are supplied by the client, not independently verified.

The browser and official MCP SDK integration are tested. Individual Codex, Claude Code and Cursor full hosting journeys remain release acceptance items; these guides are setup instructions, not a claim that every client has passed. There is no advertised one-click installation.

Codex

Add the remote server using the CLI below. If it doesn’t start authorization automatically, run codex mcp login macserve. For the full hosting workflow, request platform:read,ssh:write,connections:write,domains:write using the login command’s --scopes option, then approve those capabilities and the selected Mac in the browser. Your client may apply its own tool approval policy.

codex mcp add macserve --url https://panel.macserve.io/mcp
codex mcp login macserve --scopes platform:read,ssh:write,connections:write,domains:write

Official Codex MCP documentation

Claude Code

Add the HTTP server, open Claude Code and run /mcp to authenticate. Follow the browser approval and return to the client. A new capability may require another authorization. Use the MCP menu to inspect connection errors; do not paste MacServe passwords or private keys into chat.

claude mcp add --transport http macserve https://panel.macserve.io/mcp

Official Claude Code MCP guide

Cursor

Add this entry to your project’s .cursor/mcp.json, preserving other servers. Open Cursor’s MCP settings and connect/authenticate MacServe. Approve selected Macs and capabilities in the browser. A client-side tool disable or approval setting is separate from your MacServe grant.

{
  "mcpServers": {
    "macserve": { "url": "https://panel.macserve.io/mcp" }
  }
}

Official Cursor MCP guide

What the assistant can do

MacServe turns your Mac into a server. The assistant can guide setup from the beginning, including downloading and installing the app with your authorization when it can execute on the intended Mac. Otherwise it gives you the local installation steps. You still approve account linking, resources and macOS prompts. Once linked, MCP manages approved Macs, public connections, domains, resources, SSH public keys and permitted billing/account activity.

Your agent keeps its private SSH key and uses its own SSH client for Linux software installation on your server. DNS-provider changes, OS consent, purchases and sensitive identity changes are separate user handoffs. New Macs are never added to an existing grant automatically. Connections can be paused, archived and restored, but archived ports remain reserved. Read the account profile to check email availability; any password or inbox confirmation stays in the secure browser.

Edit or revoke access

Open AI access in the panel to see permissions, approved Macs, last use, expiry and recorded activity. Choose Edit access, select the Macs and capabilities, then Save access changes. The hosting-and-server-management preset includes SSH, publishing, domains, restart and CPU/RAM changes; account and billing permissions stay separate. Changes apply to the existing authorization on the next call, without another sign-in while it remains valid.

A reinstalled or newly linked Mac must be selected explicitly. Removing a Mac or permission blocks future platform operations. Revoke platform access invalidates the grant. Neither action removes installed SSH keys or ends existing SSH sessions; manage keys separately under the Mac’s SSH settings.

New access tokens last up to one hour. The client should renew them automatically using its rotating refresh token. Do not run multiple clients that share and race the same stored refresh token. If tools do not appear after first authorization, reconnect in the client or start a new conversation if that client requires it; do not launch a nested AI process for every call.

Agent workflow and tool reference