MCP 服务器

tokencost-mcp-server

An MCP (Model Context Protocol) server that provides real-time LLM token pricing data for 60+ AI models across 15 providers.

README

TokenCost MCP Server

An MCP (Model Context Protocol) server that provides real-time LLM token pricing data for 60+ AI models across 15 providers.

Query, compare, and estimate costs for models from OpenAI, Anthropic, Google, Meta, xAI, Mistral, DeepSeek, and more — directly from your AI assistant.

Built by TokenCost — the free LLM token cost calculator.

Tools

Tool	Description
`tokencost_get_model_pricing`	Get pricing for a specific model
`tokencost_compare_models`	Side-by-side pricing comparison
`tokencost_estimate_cost`	Calculate cost for given token counts
`tokencost_find_cheapest`	Find cheapest models with filters
`tokencost_list_models`	List all available models
`tokencost_list_providers`	List all providers with pricing ranges

Quick Start

Claude Desktop / Cursor / Windsurf

Add to your MCP config:

{
  "mcpServers": {
    "tokencost": {
      "command": "npx",
      "args": ["-y", "tokencost-mcp-server"]
    }
  }
}

From Source

git clone https://github.com/ankit-aglawe/tokencost-mcp-server
cd tokencost-mcp-server
npm install
npm run build
npm start

Example Usage

"How much would it cost to process 1M input tokens with GPT-5?"

→ Uses tokencost_estimate_cost with model="gpt-5", input_tokens=1000000, output_tokens=0

"Compare Claude Sonnet 4.6 vs GPT-5 vs Gemini 3 Pro pricing"

→ Uses tokencost_compare_models with ["claude-sonnet-4.6", "gpt-5", "gemini-3-pro"]

"What's the cheapest model with at least 200K context?"

→ Uses tokencost_find_cheapest with min_context=200000

Supported Providers

OpenAI, Anthropic, Google, xAI, Meta, Mistral, DeepSeek, Alibaba (Qwen), Amazon (Nova), NVIDIA, Cohere, Perplexity, Moonshot (Kimi), Zhipu (GLM), MiniMax

Pricing Data

Pricing is kept accurate and up to date by the TokenCost team. We track official provider announcements and update pricing as soon as changes are published — new models, price cuts, and deprecations are reflected within days.

If you notice outdated pricing or a missing model, open an issue and we'll get it updated.