Auto mode repeatedly prompts for read-only MCP calls; permission context and persistent authorization

Summary

I’m using Rowboat in regular chat with Auto mode enabled and Microsoft’s Playwright MCP extension connected to Microsoft Edge.

Browser connectivity works, but routine read-only MCP calls still require manual approval. This prevents an otherwise authorized browser workflow from running unattended.

I’m looking for either:

• Guidance on an existing supported way to authorize a defined MCP workflow persistently; or
• An improvement to Auto mode so task authorization survives long conversations and scheduled execution.

Environment

• OS: Windows
• Installed application directory indicates version: 1.0.2
• Mode: Regular chat, Auto permissions enabled; Code Mode off
• Integration: @playwright/mcp with --extension --browser msedge
• Browser: Microsoft Edge with the official Playwright extension
• Existing browser sessions are successfully reused

The MCP package was configured using @latest; I have not recorded its resolved version.

Steps to reproduce

1. Configure Playwright MCP and connect it to Edge.
2. Enable Auto permissions in a regular Rowboat chat.
3. Authorize the assistant to perform browser capability testing.
4. Ask it to list connected browser tabs or read basic browser/page status.
5. Continue the task across several conversation turns.

Expected behavior

Routine actions clearly within the owner-authorized task should run automatically.

If persistent MCP authorization is supported, I’d expect a discoverable way to configure its scope, see that it is active, and revoke it.

Actual behavior

Rowboat repeatedly displays Allow/Deny prompts for MCP calls, including:

• browser_tabs with action: "list"
• browser_evaluate running this read-only check:

() => ({
  edge: navigator.userAgent.includes('Edg/'),
  readyState: document.readyState
})