prompt-diet-mcp

Trim prompts before you pay for them.

A single MCP server you can wire into Claude Code, Cursor, or any MCP-aware client that tells you:

1. How many tokens this prompt actually costs (English + Korean aware).
2. Whether Anthropic prompt caching will work on it, and what to change if not.
3. What the monthly bill looks like at N calls/day with and without caching.
4. Whether your prompt refactor actually saved money.

No LLM call, sub-millisecond, runs over stdio.

Tools

• estimate_tokens — chars, words, approx tokens
• diet_prompt — strip filler, dedupe, normalize whitespace → trimmed prompt + suggestions
• cache_readiness — score 0-100 + concrete fix list for Anthropic 5-min cache
• simulate_cost — monthly cost across Opus / Sonnet / Haiku / GPT / Gemini with cache_hit_rate knob
• compare_prompts — A/B refactor verdict (cheaper / more cache-friendly / fewer tokens)

Install (Claude Code)

// ~/.config/claude/mcp.json
{
  "mcpServers": {
    "prompt-diet": {
      "command": "npx",
      "args": ["-y", "prompt-diet-mcp"]
    }
  }
}

Pricing

Free forever as MCP. Pro $9/mo on Gumroad for the hosted dashboard (team-wide budget tracking, slack webhook on threshold).

Contact: [email protected]