ATOM MCP Server

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

Website · About ATOM · Methodology · ATOM MCP Pro

---

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:

Category	Examples	What It Answers
Modality	Text, Multimodal, Image, Audio, Video, Voice, Embeddings	What does this type of inference cost?
Channel	Model Developers, Cloud Marketplaces, Inference Platforms, Neoclouds	Where should you buy: direct, marketplace, platform, or neocloud?
Tier	Frontier, Budget, Mid-Tier, Reasoning	What is the premium for capability?
Special	Open-Source	How much cheaper is open-weight inference?

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

---

Tools

Two complementary intelligence layers. Pricing tells you what inference costs; Model Intelligence tells you what you are paying for.

Tool	Tier	Description
list_vendors	Free	All tracked vendors with country, region, channel type, and pricing page URLs
get_kpis	Free	9 pricing KPIs: output premium, caching discount, open-source discount, context window cost, model size spread, reasoning premium, platform discount, neocloud discount, caching availability
get_model_intelligence	Free	6 capability KPIs from model metadata: reasoning tier share, long-context saturation, frontier context ceiling, output ceiling spread, training cutoff lag, vendor modality breadth
get_index_benchmarks	Free	AIPI price benchmarks across all indexes
get_market_stats	Tiered	Aggregate market intelligence: medians, quartiles, distributions, modality breakdown
search_models	Tiered	Multi-filter search: modality, vendor, creator, open-source, price range, context window, parameters
get_model_detail	Tiered	Full specs and pricing across all vendors for a single model
compare_prices	Tiered	Cross-vendor price comparison for a model or model family
get_vendor_catalog	Tiered	Complete catalog for a specific vendor

---

Pricing Tiers

	ATOM MCP (Free)	ATOM MCP Pro
Vendors, KPIs, Model Intelligence, AIPI indexes	Full data	Full data
Market stats	Aggregates only	Vendor-level breakdown
Model search and comparison	Counts and price ranges	Full granular SKU data
Model detail	Specs only	Per-vendor pricing
Vendor catalog	Summary only	Full SKU listing

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

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?"
• "How stale is the median AI model's training data?"
• "What share of models ship with long context, and how rare are reasoning models?"
• "Give me a market overview of AI inference pricing"

---

Environment Variables

Variable	Required	Description
SUPABASE_URL	Yes	Supabase project URL
SUPABASE_ANON_KEY	Yes	Supabase anonymous/public key
ATOM_API_KEYS	No	Comma-separated valid API keys for paid tier
TRANSPORT	No	stdio (default) or http
PORT	No	HTTP port (default 3000)

---

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

---

ATOM · The Global Price Benchmark for AI Inference.