Skip to main content
MCP lets an agent discover Access tools and call them using an agent profile’s credential. Access supports stdio and Streamable HTTP for agent connections.

​
Choose a transport

Under Access Channels → MCP, enable MCP and the chosen transport. Turning off the overall MCP switch blocks both transports. CLI and Mobile API are controlled separately. Use stdio or HTTP bound to 127.0.0.1 for a client on this Mac. For another device, use an authenticated, encrypted private VPN between the devices and connect through the Mac’s VPN address so the traffic uses that encrypted route. Plain HTTP alone does not protect traffic or bearer tokens in transit, even on a network you trust. Do not expose the endpoint directly to the public internet. Follow the background-service guidance when using HTTP.

​
Connect and verify

Follow Connect your first agent. Use the configuration generated by your installed Access version instead of copying another user’s executable path or credential. Refresh the client’s tool discovery after updates or grant changes, then try a bounded read and inspect Audit Log. Tool discovery does not grant access by itself. Calls still check the credential, profile, allowed operation, approval policy, and rate limits.

​
Connecting another MCP server through Access

An upstream MCP connection is a separate direction: Access connects to another app or service and exposes its reviewed tools to your agents. The server can run locally over stdio or use Streamable HTTP. Discover its tools, review their names and schemas, and explicitly enable the ones each profile needs. New and changed tools require review. Access can filter tools and apply approval policies; restrictions within a tool also depend on the upstream service. The upstream connector supports tool listing and calls. It does not expose legacy HTTP+SSE, resources, prompts, sampling, elicitation, or task-required tools. See Connection types and Supported protocols.