Routes your AI agent's API calls to the cheapest available service for each task. Instead of hardcoding providers, your agent calls generic capabilities like geocoding or translation, and Hebline picks between free options (Nominatim, MyMemory) and paid ones (Google Maps, DeepL) based on weighted scoring. Supports BYOK for paid services via environment variables. Logs every call with cost and latency to a local JSONL file. Exposes three MCP tools: execute to route and call, compare to preview available services, and categories to list supported capabilities. Currently covers nine categories including LLMs, geocoding, translation, web scraping, and weather. Roadmap includes Hebbian learning to optimize routing based on historical performance.
claude mcp add --transport stdio hebline-mcp-server -- npx -y @hebline.ai/mcp-server