Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

What an MCP server's tool list costs: we counted the tokens of 34,374 of them

Description and use

MCP server tool lists can reach hundreds of thousands of tokens, inflating request costs. Prompt caching and on-demand definition loading can cut expenses by up to 95 %.

Problem

Clients pay for sending large tool lists with every model request, inflating token usage and cost.

Outcome

Typical MCP server tool lists average 1,272 tokens (median) but can reach over 500,000 tokens, causing significant cost per request. Strategies like prompt caching and on-demand loading can reduce costs by up to 95%.

Workflow

  • Count tokens of each tool list using o200k_base tokenizer.
  • Calculate tokens per server and per tool.
  • Estimate cost per 1,000 requests for different model prices.
  • Provide a counter tool to measure a specific server's list.
  • Offer guidance on reducing token usage.

Evidence

Under our survey of 12,915 MCP servers, a reader did the arithmetic we had not: 182,157 tool definitions over 7,656 servers is about 24 tools each, and a client resends every one of them with each model request.

Source

The list rides along with every request, so its cost is tokens × price × requests.

Source

Prompt caching. If the client keeps the tool list identical and early in the prompt, a cached read costs about a twentieth of the input price.

Source

We put the counter on a free page: https://modelmarket.dev/tools/tool-list-cost

Source

  • Record date: 2026-10-19T18:41:05.760Z