Llm Prices Data

What the romanshumy/llm-prices-data MCP server does

The romanshumy/llm-prices-data MCP server exposes current LLM API pricing and model metadata through MCP tools. Its dataset covers 200 current models from 32 providers, including major vendors such as OpenAI, Anthropic, Google, Mistral, xAI, DeepSeek, Meta, Groq, and Together.

Model records include input, output, and cached-token prices per million tokens, context-window size, modality, category, release date, status, and a source URL. Provider records include descriptions and links to official pricing pages. The underlying project also publishes price history and the LLM Price Index, although the exposed tools focus on querying model and provider data.

How it works

The server retrieves live data maintained by modelpricewatch.com. Prices are checked against official provider pages and an independent reference on a recurring schedule; the README describes daily verification, with top models reviewed weekly. Results are intended for questions such as the cost of a specific model, the models available from a provider, or which options are cheapest for a selected pricing measure.

The romanshumy/llm-prices-data MCP server can be launched locally with the package command shown below. The repository also documents an HTTP endpoint at https://modelpricewatch.com/mcp, plus unauthenticated JSON API endpoints for the model and price-history datasets.

Setup and configuration

Run the local MCP process with:

npx -y @modelpricewatch/mcp

No API key or other credential is required. The README identifies Claude and Cursor as clients that can query the MCP server. The hosted MCP endpoint is available at https://modelpricewatch.com/mcp; the provided material does not specify client configuration details for that endpoint.

Tools and capabilities

• search_models searches by model name, provider, or model ID and returns current prices, context limits, modality, and category.
• get_model_pricing returns detailed pricing and capability information for one model, including cached-token price, blended cost, release date, and the modelpricewatch.com URL.
• compare_models compares two to five models by cost, context window, and capabilities, then identifies the lowest-cost option for input, output, and a typical blended workload.
• cheapest_models ranks models by input price, output price, or blended cost. Generic rankings cover generative text models; a category can be supplied for a narrower pool such as embeddings.
• list_providers returns tracked providers, short descriptions, and official pricing-page links.

Limitations and notes

The romanshumy/llm-prices-data results reflect the latest available refresh and should be checked against the linked provider page for purchase or deployment decisions. Generic cheapest-model results exclude embeddings, OCR, and realtime models because those categories price different workloads. Historical data contains both verified captures and records reconstructed from LiteLLM repository history, and some models have shorter timelines while the historical backfill is being audited.

The dataset is licensed CC-BY-4.0 and requires attribution to modelpricewatch.com. The MCP bridge code is MIT-licensed.