Even with “Run Everything (Unsandboxed)” and “Legacy Terminal Tool” enabled, chrome-devtools MCP still asks for confirmation every time

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Hello Cursor team,

I have already configured the agent with Auto-Run Mode → “Run Everything (Unsandboxed)”, and disabled browser/file/dotfile/external-file protections.
I also enabled Settings → Chat → Inline Editing & Terminal → “Legacy Terminal Tool”.

However, when I use chrome-devtools MCP for automated debugging, Cursor still prompts me for confirmation every time (Run / Skip).

Here’s the test scenario:
Prompt →

http://localhost:5173/assets Please test whether the connect-wallet button in the top navigation bar works normally.”
Request ID →
8c859ce8-c8da-4c64-9acf-6e5efc2e9f3c

Even though “Unsandboxed” mode and “Legacy Terminal Tool” are both enabled, the MCP request still requires manual approval on every step.

Steps to Reproduce

1.	Enable “Run Everything (Unsandboxed)” under Settings → Agent → Auto-Run.
2.	Disable all protections (Browser, File Deletion, Dotfile, External File).
3.	Enable “Legacy Terminal Tool” under Settings → Chat → Inline Editing & Terminal.
4.	Run a local MCP command via chrome-devtools, e.g. request to http://localhost:5173/assets.
5.	Observe that Cursor still asks for confirmation (Run / Skip).

Expected Behavior

When both “Run Everything (Unsandboxed)” and “Legacy Terminal Tool” are active, MCP and terminal requests (especially local ones) should execute automatically without manual confirmation.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

•	Cursor version: 2.0 (latest)
•	OS: [macOS ]
•	Plan: Ultra Member

For AI issues: add Request ID with privacy disabled

Request ID →
8c859ce8-c8da-4c64-9acf-6e5efc2e9f3c

Additional Information

I’m also an Ultra member, and I hope this issue can be prioritized since it significantly impacts the workflow efficiency for advanced users.

Does this stop you from using Cursor

Yes - Cursor is unusable

Hey, thanks for the report. This is a known bug affecting the “Run Everything” mode with MCP tools. The team is already working on a fix.

Temporary workaround (works for most users):

  1. Fully quit Cursor
  2. Run this community script to reset the “stuck” flags in the database: MCP Allowlist doesn't work, also can't be edited - #14 by primate
  3. Relaunch Cursor

After that, auto-run should work correctly with the chrome-devtools MCP. Several users have confirmed this resolves the issue.

Let us know if the workaround helped.