//beforeyouship — LLM cost modeling from your editor

This server connects Claude directly to beforeyouship's LLM cost modeling engine. It exposes three tools: estimate_cost runs a full architectural forecast with naive, realistic, and worst-case monthly spend across 18 models (6 free, full catalog with a Pro key), accounting for retries, prompt caching, batch discounts, and infrastructure overhead. get_model_prices returns current per-million-token rates with context windows. list_archetypes surfaces seven preset patterns like RAG pipelines and multi-step agents. Reach for this during the design phase when you need to compare Anthropic, OpenAI, Gemini, and DeepSeek costs before writing code. The underlying model uses calibrated tokenization ratios, retry cascades for chained calls, and adjustable cache hit rates. Works in demo mode without credentials.