atom-mcp-server

What This Is

ATOM MCP Server lets any MCP-compatible AI agent (Claude, GPT, Cursor, Windsurf, VS Code Copilot) query live AI inference pricing data programmatically. Built on financial index methodology comparable to S&P Dow Jones, MSCI, and Bloomberg, the ATOM Inference Price Index (AIPI) is the first independent benchmark for the AI inference market.

Ask your AI assistant a question like "What's the cheapest way to run GPT-4o?" and it calls ATOM's tools behind the scenes, returning a data-backed answer pulled from live coverage across the global vendor landscape.

AIPI Indexes

The AIPI index family spans four categories, all calculated weekly using chained matched-model methodology to eliminate composition bias:

All indexes are global (GLB) and reported across three pricing directions: input, cached input, output.

Tools

Pricing Tiers

Free tier (no API key): enough to understand the market through counts, ranges, distributions, and benchmarks.

ATOM MCP Pro: full granular data across every vendor, model, price, and spec. Subscribe at a7om.com/mcp.

Quick Start

Option 1: Remote URL for Claude.ai and Claude Desktop (recommended)

No install required. Connect directly to ATOM's hosted server.

Claude.ai (web): Settings → Connectors → Add custom connector

Name: ATOM Pricing Intelligence
URL:  https://atom-mcp-server-production.up.railway.app/mcp

Claude Desktop: Settings → Developer → Edit Config

{
  "mcpServers": {
    "atom-pricing": {
      "url": "https://atom-mcp-server-production.up.railway.app/mcp"
    }
  }
}

Note: Remote URL support requires a recent Claude Desktop version. If it does not work, use the npx method below.

Claude Desktop (via npx proxy):

{
  "mcpServers": {
    "atom-pricing": {
      "command": "npx",
      "args": ["mcp-remote", "https://atom-mcp-server-production.up.railway.app/mcp"]
    }
  }
}

Option 2: Local (stdio) for Cursor, Windsurf, and similar clients

git clone https://github.com/A7OM-AI/atom-mcp-server.git
cd atom-mcp-server
npm install && npm run build

Add to your MCP client config:

{
  "mcpServers": {
    "atom-pricing": {
      "command": "node",
      "args": ["/path/to/atom-mcp-server/dist/index.js"],
      "env": {
        "SUPABASE_URL": "https://jonncmzxvxzwyaznokba.supabase.co",
        "SUPABASE_ANON_KEY": "your-anon-key"
      }
    }
  }
}

Option 3: Deploy your own (Railway)

Set environment variables in Railway dashboard:

• SUPABASE_URL

• SUPABASE_ANON_KEY

• ATOM_API_KEYS (comma-separated, for paid tier validation)

• TRANSPORT=http

Example Queries

Once connected, ask your AI assistant in natural language:

• "What's the cheapest way to run GPT-4o?"

• "Compare Claude Sonnet 4.5 pricing across all vendors"

• "Find open-source text models under $0.50 per million tokens"

• "Show me Google's full model catalog"

• "What are the AIPI benchmark prices for text inference?"

• "How do neocloud prices compare to cloud marketplaces?"

• "How much cheaper is open-source inference?"

• "Give me a market overview of AI inference pricing"

Environment Variables

Tech Stack

• TypeScript / Node.js

• MCP SDK (@modelcontextprotocol/sdk)

• Supabase (PostgreSQL) via REST API

• Express (HTTP transport)

• Zod (schema validation)

About ATOM

The Global Price Benchmark for AI Inference. Independent pricing intelligence for developers, analysts, and infrastructure buyers. Transparent methodology, deterministic indexing, weekly market intelligence across the global AI inference market.

ATOM was founded in 2025 by Stamos Kanellakis. The platform is built on financial index methodology comparable to S&P Dow Jones, MSCI, and Bloomberg, and applies that discipline to a market that previously had no pricing benchmark of its own.

Products: ATOM MCP · ATOM Terminal · ATOM Feed

License

MIT