Labsco
MCP SERVER

Artificial Analysis

by davidhariri ↗

Look up what a model costs, how fast it runs and how it scores on benchmarks, in the same session where you are deciding which one to call.

Model Routing, Multi-Model Consultation & Cost Control✓ Verified
Summary
Two tools, one job: turning "which model should this run on" into numbers comparable in the same breath as the code that will call it.

The shape is list-then-detail. list_models narrows by creator and orders by price, speed or benchmark score; get_model fills in the full record for whichever candidate survived that cut. Nothing here calls a model — these are Artificial Analysis's published measurements, so the numbers are as current as that source and no more.

What it is

A read-only lookup over Artificial Analysis's model data, with a list-and-filter tool and a detail tool for a single named model.

What you get
  • list_models returns available LLM models with pricing, speed and benchmark data, filtered by creator — OpenAI, Anthropic, Google among them.
  • sort_by, sort_order and limit order that list by price, speed or benchmark score and cut it to the length you want to read.
  • get_model returns a named model's pricing split into input, output and blended per 1M tokens, which is the shape a cost estimate actually needs.
  • get_model also returns speed as tokens/sec and TTFT, alongside benchmark scores including Intelligence Index, Coding Index, MMLU-Pro and GPQA.
Requirements

An Artificial Analysis API key, supplied as AA_API_KEY.

Setup effort

One command plus a key — claude mcp add artificial-analysis -e AA_API_KEY=your-key -- npx -y artificial-analysis-mcp, then supply credentials