[FEATURE] Request: Transparent Context Usage & Tool Loading Optimization

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Title: Feature Request: Transparent Context Usage & Tool Loading Optimization

### Problem

Claude Code advertises 200K context window, but the actual usable capacity is significantly lower due to system overhead:

| Component | Tokens | % of Total |
|-----------|--------|------------|
| System tools | 20.3K | 10.1% |
| System prompt | 11.9K | 6.0% |
| Total system overhead | ~32K | ~16% |
| Actual usable context | ~168K | ~84% |

This means users lose approximately 16% of their context window before writing a single message.

### Environment

• Claude Code version: (run claude --version)
• Model: claude-opus-4-5-20251101
• Context observed via /context command

---

Thank you for considering this feedback. Claude Code is a powerful tool, and transparent resource usage would improve user trust and experience.

Proposed Solution

### Proposed Improvements

1. Lazy Tool Loading: Only load tool definitions when first invoked or explicitly enabled
2. Transparent Capacity Display: Show "168K usable / 200K total" instead of just "200K"
3. Tool Selection Option: Allow users to disable unused built-in tools (e.g., NotebookEdit, LSP if not needed)
4. Compressed Tool Definitions: Consider shorter schemas or server-side tool resolution

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

### Current Behavior

• All built-in tools are loaded on every request, regardless of whether they're used
• The /context command shows usage, but advertised "200K" creates misleading expectations
• Adding MCP servers further reduces available context (e.g., Serena adds 8.5K)

Additional Context

### Impact

For users working with large codebases or long conversations, 32K of pre-consumed context is substantial. This is especially impactful when:

• Analyzing large files
• Maintaining long debugging sessions
• Using multiple MCP servers simultaneously