Documentation
Set up Captain Relay, enrol a computer, and connect an AI assistant through the secure MCP endpoint.
https://mcp.captainrelay.com/mcp
Authentication uses the Captain Relay sign-in flow. Do not share access tokens, device credentials or enrolment links.
Quick start
Open the Captain Relay app and complete authentication.
Open Devices, choose the operating system and generate the short-lived setup command. Run it only on a computer you own or are authorised to administer.
Captain Relay Agent connects outbound to the service. No public inbound control port is required on the computer.
For each capability, choose whether Captain Relay should allow it, deny it or require human approval.
Add Captain Relay as a remote MCP connection using the production endpoint above, then sign in when prompted.
Safety model
The assistant does not receive unrestricted machine access. It sees the capabilities reported by the enrolled Captain Relay Agent and every requested action is checked against server-side ownership and policy controls.
Human approvals
If a policy requires approval, the action is queued but does not execute. After the decision, the client can retrieve the approval result without resubmitting the original action.
Persistent jobs
Builds, tests, installs and other long-running work should use persistent execution jobs. Jobs retain lifecycle state so the client can poll progress, cancel work or retrieve results after reconnecting.
Computer use
On supported computers, screen capture, window discovery and pointer/keyboard input are separate explicit capabilities and can be independently controlled by policy.
AI clients
Captain Relay is designed around the open Model Context Protocol. Directory listings for supported AI clients are part of the public launch process; until a directory listing is published, use the production MCP endpoint only in clients that support custom remote MCP connections.