llm-cost-estimator-mcp

An MCP (Model Context Protocol) server that gives Claude, Cursor, or any MCP client the ability to count tokens and estimate LLM costs across GPT-4o, Claude, Gemini, DeepSeek, Mistral, Llama, Grok, and 25+ models — locally, with no API key.

It reuses the exact same tokenizer + pricing + estimator core as the LLM Cost & Token Estimator VS Code extension in this repo.

Tools

| Tool | Description | | --- | --- | | estimate_cost | Token count + cost across multiple models for a piece of text. Returns a Markdown table and structured JSON. Args: text, models?, outputTokens?, currency?. | | count_tokens | Token count for one model. Args: text, model?. OpenAI is exact (tiktoken); others are flagged estimates. | | list_models | All model keys with provider, prices (USD/1M), and context window. |

⚠️ Prices are representative placeholders — verify against each provider's official pricing page. OpenAI counts are exact; all other providers are approximated via an OpenAI encoding and flagged as estimates.

Build (from source)

The server reuses the extension's core in ../src (whose files import js-tiktoken), so install the repo-root dependencies first:

npm install            # at the repo ROOT, once (provides the reused core's deps)
cd mcp-server
npm install
npm run build          # -> dist/index.js
npm test               # builds, then runs a real stdio MCP handshake

End users don't need this — they just npx -y llm-cost-estimator-mcp once it's published (the published dist/index.js is fully self-contained).

Connect it

The server speaks MCP over stdio, so any client launches it as a subprocess. Use the absolute path to mcp-server/dist/index.js.

Claude Desktop

Edit claude_desktop_config.json (Windows: %APPDATA%\Claude\claude_desktop_config.json, macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "llm-cost-estimator": {
      "command": "node",
      "args": ["C:/path/to/llm-cost-estimator/mcp-server/dist/index.js"]
    }
  }
}

Restart Claude Desktop, then ask: "Estimate the cost of this prompt on GPT-4o vs Claude Sonnet vs Gemini."

Claude Code (CLI)

claude mcp add llm-cost-estimator -- node /abs/path/to/mcp-server/dist/index.js

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "llm-cost-estimator": {
      "command": "node",
      "args": ["/abs/path/to/mcp-server/dist/index.js"]
    }
  }
}

After publishing to npm (optional)

Once this package is published, clients can launch it without a local build:

{
  "mcpServers": {
    "llm-cost-estimator": { "command": "npx", "args": ["-y", "llm-cost-estimator-mcp"] }
  }
}

License

MIT — see ../LICENSE.