Artificial Analysis
@davidhariri
I built this to give my AI assistants the latest information on model performance benchmarks provided by https://artificialanalysis.ai to make informed decisions on model choice.
Overview
What is Artificial Analysis?
An MCP (Model Context Protocol) server that provides real-time LLM model pricing, speed metrics, and benchmark scores from Artificial Analysis. It is designed for developers and AI assistants like Claude to programmatically query and compare 300+ models.
How to use Artificial Analysis?
Install via npx with your API key set as the AA_API_KEY environment variable, or add the server to your Claude configuration file (settings.json). Use tools like list_models and get_model to fetch and filter model data by creator, metric, or benchmark.
Key features of Artificial Analysis
- List and filter 300+ LLM models by creator and metric
- Get real-time pricing (input/output/blended per 1M tokens)
- Compare speed metrics (tokens/sec, time to first token)
- Access benchmark scores (Intelligence Index, Coding Index, etc.)
- Sort models by any metric (price, speed, benchmarks, release date)
Use cases of Artificial Analysis
- Find the cheapest models with high intelligence scores
- Compare speed of models from different providers
- Check benchmark scores of a specific model like GPT-4o
- Identify top 5 fastest models for a new project
- Get detailed pricing and metrics for Claude 4.5 Sonnet
FAQ from Artificial Analysis
What API key is required?
An Artificial Analysis API key (AA_API_KEY) is required. You can obtain one at artificialanalysis.ai.
How do I install the server?
Install via npx: npx -y artificial-analysis-mcp with the AA_API_KEY environment variable. Alternatively, add it to Claude’s MCP configuration in ~/.claude/settings.json.
What model data is available?
Each model includes pricing (input/output/blended per 1M tokens), speed (output tokens/sec, time to first token), and benchmarks (Intelligence Index, Coding Index, Math Index, MMLU-Pro, GPQA, LiveCodeBench, and more).
Can I filter or sort models?
Yes. The list_models tool supports filtering by creator (e.g., "OpenAI") and sorting by fields like price, speed, intelligence index, or release date.
Is there a limit on results?
Yes, the list_models tool accepts an optional limit parameter to cap the number of returned models.
