claude-cost-audit-mcp
Exact Claude API cost calculation with correct prompt-cache economics (1.25x/2x write, 0.1x read) and cache break-even analysis, plus a scanner that catches the single most common Claude-cost mistake — estimating tokens with OpenAI's tiktoken, which Anthropic's own docs say undercounts Claude by 15-20%.
How to pay
Subscribe
$7/month
Predictable monthly cost with included usage. Best for steady, high-volume traffic.
- Unlimited tools within plan limits
- One API key, billed once a month
- Cancel any time
Deploys three tools for exact Claude API cost accounting: calculate_message_cost computes real cost per request using Anthropic's current per-model pricing, pulled from the live claude-api skill rather than recalled from training data (which would miss things like a model's time-limited introductory rate); check_cache_breakeven applies the correct prompt-cache math — 1.25x write cost at the 5-minute TTL, 2x write cost at the 1-hour TTL, 0.1x read cost — and tells you the exact number of reuses needed to break even (2 at 5-minute TTL, 3 at 1-hour TTL); audit_tokenizer_usage flags the single most common Claude-cost mistake, estimating tokens with OpenAI's tiktoken, which undercounts Claude token counts by roughly 15-20% per Anthropic's own documentation. Free tier included, no API key required.