Every MCP client ships the model the full JSON schema of every tool before
you ask a single question. This calculator estimates that bill for your own setup.
🔒 runs entirely in your browser — nothing is sent anywhere
Your setup
Three setups we measured
end to end; each loads its own measured tokens-per-tool rate.
Count them with mcptoon manifest --compact.
282 is the measured mean of a 255-tool / 50-server listing
(71,929 tokens, tiktoken cl100k_base). Descriptions tuned for LLM
comprehension push this up; terse servers pull it down.
The assumption changes the answer by 20×, so it is stated, not buried.
What you pay for discovery
—
tokens per listing, sent before your first question
—
of that never has to enter the window (— tokens instead)