Summary
Model pricing is the fact an assistant is most confidently wrong about.
Prices change and training data does not, so a model asked what something costs answers from memory and is out of date. These calls put a maintained price table in front of it instead.
What it is
A pricing lookup for large language models across the providers the vendor tracks. It runs locally from npx and answers from a price table the vendor maintains.
What you get
- tokencost_get_model_pricing — the price for one named model
- tokencost_compare_models — several models priced side by side
- tokencost_estimate_cost — the cost of the input and output token counts you supply
- tokencost_find_cheapest — the cheapest models that pass filters such as a minimum context window
- tokencost_list_models and tokencost_list_providers — every model it knows, and every provider with its pricing range
Requirements
Npx on your PATH. No account and no key.
Setup effort
One command — npx -y tokencost-mcp-server