context-auditlisted

# Context Audit Audit your Claude Code setup for token waste and context bloat. You'll get a scored health report with prioritized fixes. **Why this matters**: Each MCP server burns ~15–20K tokens per turn. A 500-line CLAUDE.md floods every prompt with noise. Redundant skills inflate the system prompt silently. Together, these cause Claude to ignore rules, degrade mid-session, and exhaust usage limits early. --- ## Step 1 — Gather /context data **Do NOT proceed to Step 2 until the user has run `/context` and shared the output.** Ask the user: > Run `/context` in your terminal and paste the full output here. The context breakdown is required to determine what to audit and in what order. The `/context` output will show token counts broken down by: - System prompt - Tools (MCP servers) - Files in context - Conversation history If the user cannot run `/context`, proceed with a filesystem-only audit and note that scores may be incomplete. --- ## Step 2 — Systematic audit Work through each area below. Read the actual files — do not assume. ### 2A — MCP Servers Read `~/.claude/settings.json` and any project-level `.claude/settings.json`. For each MCP server found: - Note name and transport type - Check if a zero-cost CLI alternative exists (e.g., `gh` replaces GitHub MCP, `neonctl` replaces Neon MCP) - Flag servers that have been connected but unused in recent sessions (check if tools appear in `/context` tool list) **Token cost**: ~15,000–20,000 tokens per server pe