VS Code: MCP permission prompts show only the tool name, not what the call does

What happens

When an MCP tool call needs permission in the VS Code extension panel, the prompt is only:

Do you want to proceed with mcp__claude_ai_Linear__list_issues?

It doesn't show the arguments, a description, or whether the call reads or writes. I can't make an informed decision, so I either approve blindly or decline and ask what it was.

Why it matters

• Tool names are machine identifiers (mcp__<server>__<tool>). You can't tell from the name alone whether a call only reads or makes changes: compare list_issues, save_issue and merge_diff.
• Bash prompts show a plain-language description the model writes for each command. MCP calls, which often reach external services, get less context than a local shell command.
• The workaround today: Claude asks a plain-language question first, and then the tool-name prompt follows. It works, but it doubles the prompts.

Suggestions (any one would help)

1. Show the call's arguments in the prompt, e.g. query: "Douven", includeArchived: true
2. Show the tool's description from the MCP server's schema
3. Let the model supply a one-line description, as Bash already does
4. Use the MCP readOnlyHint / destructiveHint annotations to label a call as a read or a write

Environment: VS Code extension anthropic.claude-code 2.1.282, VS Code 1.139.0, macOS 27.0, claude.ai Linear connector