[BUG] Built-in mcp__ide__executeCode asks for permission in VS Code pop-up for every notebook cell even if already allowed

[JasonLiJT]

Environment

• Platform (select one):
  • [ ] Anthropic API
  • [ ] AWS Bedrock
  • [x] Google Vertex AI
  • [ ] Other:
• Claude CLI version: 1.0.10 (Claude Code)
• Operating System: Linux 4.18.0 (server side)
• Terminal: VS Code 1.100.2 integrated terminal, connecting to the Linux server with remote ssh

Bug Description

After Claude Code uses the mcp__ide__executeCode tool for the first time, if you choose to "allow and don't ask again", then the mcp call itself is auto-allowed. However, vscode still pops up to ask for permission to execute every new cell.

That said, mcp__ide__executeCode is still very useful even with the pop-up for every cell. Thank you for this built-in mcp tool and claude code itself!

Steps to Reproduce

1. Open a notebook and claude code side-by-side in VS Code.
2. Ask claude code to execute some code directly in the notebook to get it to use the mcp__ide__executeCode tool.
3. If it's the first time claude code uses mcp__ide__executeCode, select "allow and don't ask again".
4. vscode pops up a drop-down menu, asking for permission to execute a new code cell in the notebook. Click Execute.
5. Ask claude code to execute another cell.

Expected Behavior

The new cell is executed without any prompt asking for permission.

Actual Behavior

vscode pops up to ask for permission again.

Additional Context

Claude Code (CLI) auto-installs the Claude Code vscode extension from integrated terminal. The extension version is also 1.0.10.

[ntoshev]

+1

[ewengillies]

Agree that this slows down the overall workflow. It feels like an issue with the VSCode integration - not sure if there is a permission setting on the VSCode side of things.

As a workaround, I've been using papermill (https://github.com/nteract/papermill) to run notebooks in 'headless mode' with VSCode. Claude then just uses the normal "Bash(" tool to run the papermill <input_notebook> <output_notebook> workflow.

You have to be careful with the --kernel parameter to ensure its pointing at the correct IPython kernel for your python environment but otherwise this workflow works well.

[pdwhoward]

I have the same issue. Please make an option to auto approve!

[github-actions[bot]]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

[stefandanzl]

The issue still persists