[BUG] Claude Code should assess tool selection and prompt for plan mode before executing multi-step MCP tasks

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Title: Claude Code should assess tool selection and prompt for plan mode before executing multi-step MCP tasks

Category: UX / Behavioral Guardrail — Feature Request

Summary

Claude Code executed a multi-step Google Drive MCP workflow without pausing to assess whether the tool was appropriate, clarify scope, or offer plan mode. The result was unnecessary token consumption and a task that would have been better handled in claude.ai chat.

What Happened

User asked Claude Code to rename files in a Google Drive folder. Claude Code immediately:

Loaded Drive MCP tool schemas
Searched the folder and retrieved full content snippets
Created a backup folder and copied 4 files
Downloaded a 32KB HTML file as base64 to perform a text replacement
Total session cost: ~40,000–60,000 tokens. The same task completed in claude.ai chat with file upload would have cost a fraction of that.

The core problem: the Drive MCP has no rename function. Claude Code did not surface this constraint before executing. The workaround (download content → replace text → create new file) was expensive and unnecessary.

Root Cause

Claude Code has no built-in guardrail that:

Assesses whether a task is better suited for claude.ai chat vs. MCP tools
Prompts the user to choose plan mode vs. execute mode before multi-step tasks begin
Surfaces tool capability gaps before the first API call
Estimates and flags token cost before committing to an approach
Impact

Unnecessary token spend for the user
Irreversible side effects (files created in Drive that user must manually clean up)
Eroded trust in Claude Code as a reliable agentic tool
Suggested Fix

Before any task involving MCP tools, external APIs, or more than two sequential steps, Claude Code should automatically:

Assess tool fit — Is this editing or automation? If editing, recommend claude.ai chat explicitly.
Prompt for plan vs. execute — "Would you like me to plan first or execute immediately?" and wait for a response.
Surface constraints — Flag missing tool capabilities (e.g., no rename in Drive MCP) before any API calls.
Estimate cost — Warn if the approach requires downloading large content or many sequential calls.
This behavior should be the default, not dependent on user-configured CLAUDE.md instructions.

Workaround (User-Applied)

Added Planning Protocol and Tool Selection Protocol directly to global CLAUDE.md. This works but should not be required — it should be native behavior.

Environment

Claude Code CLI / VS Code Extension
Model: claude-sonnet-4-6
MCP: Google Drive

What Should Happen?

What Should Happen
When a user submits a request that involves MCP tools, external APIs, file operations, or more than two sequential steps, Claude Code should pause before executing and respond with a structured pre-task assessment:

1. Tool Fit Check

Claude Code should evaluate whether the task is better suited for claude.ai chat or Claude Code + MCP. If the task is document editing, content updates, or a small number of files, it should say so explicitly and recommend the cheaper path before loading any tools.

1. Plan or Execute Prompt

Claude Code should ask: "Would you like me to plan this out first, or execute immediately?" and wait for a response. It should not proceed until the user answers. Plan mode should be the recommended default for any task that touches external systems or cannot be easily undone.

1. Constraint Disclosure

Before the first API call, Claude Code should identify and surface any capability gaps in the available tools. In this case: "The Drive MCP does not have a rename function. The workaround is copy + recreate, which requires downloading file content and will consume significantly more tokens. Do you want to proceed?"

1. Cost Estimate

If the approach involves downloading large files, chaining many API calls, or generating substantial output, Claude Code should estimate the token impact and ask for confirmation before committing.

None of this should require user-configured CLAUDE.md instructions. It should be default behavior — the same discipline a senior engineer applies before touching production systems.

Error Messages/Logs

Steps to Reproduce

Steps to Reproduce
Open Claude Code with a Google Drive MCP integration configured
Submit a natural language request to rename or update files in a Drive folder — e.g., "Change the name of all material in my Drive folder from AI Exec Coach to MyBoard.ai"
Observe that Claude Code immediately begins executing:
Loads MCP tool schemas without assessing tool fit
Searches the folder and retrieves full file content snippets
Makes multiple sequential API calls without pausing
Downloads binary/large file content (HTML, docx) unnecessarily
No plan mode prompt is offered at any point
No tool capability gaps are surfaced (e.g., Drive MCP has no rename function)
No token cost estimate is provided before committing to the approach
Expected: Claude Code pauses, assesses tool fit, surfaces constraints, and asks "Plan first or execute immediately?" before any API calls are made.

Actual: Claude Code executes immediately, consuming 40,000–60,000 tokens for a task that required fewer than 5,000.

Claude Model

Sonnet (default)

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.126

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_