tokens.count

Exact token count for the OpenAI o200k_base and cl100k_base encodings, computed offline.

DataLaunch set1 unit per callnot cachedMCP: tokens_countv1.0.0

When to use it

Use to estimate prompt size or cost for OpenAI models. Counts are exact for the named encoding. Anthropic and Google tokenizers differ and are not supported yet.

Input

Field	Type	Description
`text` *	`string (1+ chars, ≤ 200000 chars)`	Text to tokenize, up to 200,000 characters.
`encoding`	`one of o200k_base, cl100k_base`	o200k_base for GPT-4o and later. cl100k_base for GPT-4 and GPT-3.5. Default: "o200k_base".

* required

Output

Returns data with: tokens, characters, encoding, exact (true for supported encodings).

Request

curl -X POST https://agentops.tools/v1/tokens.count \
  -H 'content-type: application/json' \
  -d '{"text":"Hello, world!"}'

Over MCP the tool is named tokens_count. Over A2A its skill id is tokens.count. Same input, same envelope.

Try it

Input (JSON)

Counts against your rate limit.

Sources

tiktoken encodings via js-tiktoken

Tokenization by js-tiktoken (MIT), using OpenAI encodings