Skip to content
tokencost tokencost

Token Cost

Real LLM pricing data for your AI tools — not hallucinated guesses. An MCP server that gives Claude, Cursor, and other AI assistants accurate, up-to-date model pricing from the LiteLLM community registry.
Get Started View on GitHub

See it in action

Section titled “See it in action”

tokencost demo — ask pricing questions in natural language and get instant answers

Install

Section titled “Install”

Add to your Claude Code config in one command:

Terminal window
claude mcp add tokencost-dev -- npx -y tokencost-dev

Or add the JSON config manually — see the Getting Started guide for VS Code, Cursor, Windsurf, and more.

What can it do?

Section titled “What can it do?”

Model Lookup

Look up pricing, context windows, and capabilities for any LLM. Fuzzy matching means you don’t need exact model IDs.

Cost Estimation

Estimate the cost for a given number of input and output tokens on any model. Know what you’ll spend before you spend it.

Model Comparison

Filter and compare models by provider, context window, or mode. Find the most cost-effective model for your workload.

Live Data

Pricing data sourced from the LiteLLM community registry, cached for 24 hours with on-demand refresh.

Example: batch cost planning

Section titled “Example: batch cost planning”

“How much would it cost to run Claude Opus vs Sonnet for 10K requests, averaging 2K input and 500 output tokens each?”

Claude Opus 4.6:   $450.00  (Input: $300 + Output: $150)
Claude Sonnet 4.6: $135.00  (Input: $60  + Output: $75)

Sonnet is 3.3x cheaper — saving $315 per 10K-request batch.

See more examples →