Model Context Protocol

Use AI Economy Lab from your AI assistant.

Same dataset and tools as the calculator on this site, callable from Claude Desktop, Cursor, Claude Code, or any MCP-aware client. Zero local install — point your client at the URL.

30-second install

Add this to your client config and restart. The assistant gets five tools and three resources for live LLM pricing.

Endpoint


https://aieconomylab.com/mcp

{
  "mcpServers": {
    "ai-pricing": {
      "url": "https://aieconomylab.com/mcp"
    }
  }
}

Five tools, ready in chat

Ask in plain English. The assistant calls the right tool, gets fresh pricing, answers with numbers.

estimate_cost

Per-call, monthly, and yearly cost for a workload. Splits input, cached input, and output. Applies language overhead.

Try saying

"Estimate the monthly cost of running gpt-5 with 5K input / 1K output per call, 50% cached, 10K calls a month."

compare_models

Side-by-side pricing and capabilities for 1–6 models. Useful before locking in a vendor.

Try saying

"Compare gpt-5, claude-opus-4-7, and gemini-3-1-pro side by side."

find_cheapest

Cheapest model meeting your filters: vision, tools, caching, reasoning, structured output, context window.

Try saying

"What's the cheapest model with vision and a 200K+ context window, sorted by output cost?"

list_models

Every tracked model, optionally filtered by provider or family.

Try saying

"List all OpenAI models with their input and output prices."

tokenizer_overhead

Per-language token-count multiplier for o200k_base, claude, or gemini tokenizer families. Explains why your non-English users cost more.

Try saying

"How many more tokens does Turkish content take on the o200k_base tokenizer compared to English?"

How it works

AI Economy Lab maintains an open dataset of LLM prices and tokenizer ratios. The calculator on this site, the npm-distributed stdio MCP server, and this hosted endpoint all read from the same dataset — pricing cannot drift between them. Stateless transport: a fresh server is spun up per request, no cross-request state, no per-user logging.

View dataset on GitHub

Prefer to self-host?

The npm package ships a binary that runs over stdio with no network calls. Same five tools, same dataset.

npx -y -p @aieconomylab/ai-pricing ai-pricing-mcp

FAQ

› What is MCP?

Model Context Protocol — an open standard from Anthropic that lets AI assistants call external tools. Once configured, your assistant can fetch live data, run calculations, or access services. Supported in Claude Desktop, Cursor, Claude Code, and a growing list of clients.

› Does it cost anything?

No. Hosted as part of the open-source AI Economy Lab. Free for any tool, agent, dashboard, or research project.

› What about privacy?

We don't log your queries or store API keys — there's no auth and no per-user state. Each request is stateless. Server logs only include standard request metadata for abuse protection.

› How fresh is the data?

Every pricing entry carries a verification date. We aim to verify weekly and update within 48 hours of any vendor pricing change. The dataset on GitHub is the source of truth.

› Can I self-host?

Yes. The npm package ships a binary that runs over stdio with no network calls. Same five tools, same dataset.

No MCP-aware client?

The interactive calculator gives you the same numbers in your browser — no setup.

Open the calculator