I configured a custom MCP server using Remote URL:
http://<tailscale-ip>:8000/mcp
The server runs on another machine reachable over Tailscale. OpenWorker successfully discovered and invoked its echo tool.
Before invocation, the approval card displayed:
stays on this computer
However, the invocation crossed the network from the OpenWorker host to the remote MCP server.
Expected: wording that distinguishes local-process execution from remote/self-hosted MCP execution, e.g. “connects to a remote MCP server” or similar.
Actual: remote HTTP MCP tool is labeled “stays on this computer.”
Is “stays on this computer” intended to mean something narrower, such as “not sent through OpenWorker Cloud”? If so, the current wording may be misleading because it appears to describe the tool’s execution/data boundary.
Reproduced with OpenWorker main, MCP 1.29.1, Streamable HTTP, and a remote server over Tailscale.
I configured a custom MCP server using Remote URL:
http://<tailscale-ip>:8000/mcpThe server runs on another machine reachable over Tailscale. OpenWorker successfully discovered and invoked its
echotool.Before invocation, the approval card displayed:
stays on this computerHowever, the invocation crossed the network from the OpenWorker host to the remote MCP server.
Expected: wording that distinguishes local-process execution from remote/self-hosted MCP execution, e.g. “connects to a remote MCP server” or similar.
Actual: remote HTTP MCP tool is labeled “stays on this computer.”
Is “stays on this computer” intended to mean something narrower, such as “not sent through OpenWorker Cloud”? If so, the current wording may be misleading because it appears to describe the tool’s execution/data boundary.
Reproduced with OpenWorker
main,MCP 1.29.1, Streamable HTTP, and a remote server over Tailscale.