Your tokens are already spoken for

System prompts and MCP tool definitions get billed and counted before a single message is exchanged. Paste yours in, pick a model, and see what's actually left to work with.

0% of the window already spent before you type a word
Showing example numbers — paste your own below to replace them.
~0 tokens
~0 tokens
No JSON handy? A rough list works too — this is an estimate either way, not a billed count.
Overhead before your first message 0 tokens
system prompt tools
Left for actual conversation 0 tokens

Common questions

How many tokens do MCP tools use?

Each connected MCP tool adds its name, description, and parameter schema to every request, typically 20 to 150 tokens per tool depending on how detailed its description is. With several MCP servers connected at once, this can add up to hundreds or thousands of tokens before you send a single message.

Why is my context window already full before I type anything?

Your system prompt and any connected tool definitions are sent with every request and counted against the same context window as your conversation. A long system prompt combined with several MCP tools can quietly use a meaningful share of the window before your first message is even sent.

How do I estimate system prompt token cost?

A rough rule of thumb is about 4 characters per token for plain English text, and closer to 3.3 characters per token for punctuation-heavy text like JSON tool schemas. Pasting the actual text into a calculator gives a closer estimate than guessing from word count.