UltraRAG is a framework designed to build complex RAG (Retrieval-Augmented Generation) pipelines with minimal code. It provides MCP (Model Context Protocol) integration for connecting various LLM providers including OpenAI, Qwen, DeepSeek, and supports multimodal models. The framework includes support for embedding models via sentence-transformers, model serving via vLLM, and offers a UI for pipeline visualization. It also incorporates agent capabilities as demonstrated by the AgentCPM-Report on-device writing agent.