{"name":"OnPrem World","description":"MCP server (Streamable HTTP). POST JSON-RPC 2.0 messages to this URL.","tools":[{"name":"search_hardware","description":"Find GPUs, Apple Silicon Macs and unified-memory mini PCs for running LLMs locally. Returns memory size, memory bandwidth and IDs."},{"name":"search_models","description":"Find open-weight LLMs people run locally. Returns parameters, licence, Ollama tag and IDs."},{"name":"check_fit","description":"Will a model run on a piece of hardware? Memory needed per quantization (Q4_K_M…FP16), whether it fits, and estimated tokens/s. Estimates, not benchmarks."},{"name":"what_can_run","description":"List the models a device can run, with the best quantization that fits and estimated tokens/s."},{"name":"get_details","description":"Full record for one hardware ID or model ID, including source URLs for every spec. For models, also the memory table per quantization and context length."},{"name":"gpu_prices","description":"Latest prices for GPUs, Macs and mini PCs in Taiwan (TWD) and the US (USD): each price with its shop URL and check date, price per GB of memory, and a value score (estimated tokens/s per NT$1,000 or US$100). Prices are never converted between currencies."},{"name":"latest_news","description":"Latest local-AI news (open model releases, runtime releases such as Ollama/llama.cpp/vLLM, hardware launches), newest first. Each item has a short English and Traditional Chinese summary, why it matters for running models locally, and the official source URL."}],"api":"https://onpremworld.com/openapi.json"}