Skip to main content

UltraRAG 2.0

UltraRAG 2.0 (UR-2.0) is the first lightweight framework for building Retrieval-Augmented Generation (RAG) systems based on the Model Context Protocol (MCP) architecture. It is designed to provide efficient modeling support for scientific research and exploration. UR-2.0 standardizes core RAG components—such as Retriever and Generation—into independent MCP Servers, and enables flexible integration and extension through function-level Tool interfaces. With the MCP Client serving as a process scheduler, developers can build complex reasoning pipelines featuring sequential, loop, and conditional branching structures simply by writing concise YAML files. This approach significantly lowers the technical barriers and learning curve for developing and debugging sophisticated RAG systems.

📑 Pipeline

A YAML-based process definition file used to describe the execution order and logic of component and tool calls.

🕹️ Client

The central controller of the workflow that parses the Pipeline and coordinates the execution order of Tools across different Servers.

🔧 Server

A collection of independently callable functional modules. Developers can easily add new features or modules via standardized interfaces for flexible extension and efficient composition.
In addition, UR-2.0 provides step-by-step tutorials, integrates multiple mainstream benchmark tasks and high-quality baselines, and offers a unified evaluation system and knowledge base to greatly improve development efficiency and ensure experimental reproducibility.

Why UltraRAG 2.0?

RAG systems are evolving from simple, modular pipelines into complex architectures that rely on model-driven reasoning, dynamic retrieval, and conditional decision-making. Traditional frameworks often lack the flexibility and scalability required for multi-turn interactions and adaptive information updates. Furthermore, existing open-source implementations typically lack a unified paradigm, leading to tightly coupled and inconsistent baseline modules that hinder understanding, reproducibility, and comparative research. To address these challenges, UltraRAG 2.0 is designed to offer:

🚀 Low-code construction of complex pipelines

Natively supports sequential, looping, and conditional reasoning structures. Developers can implement iterative RAG workflows by writing simple YAML files with just a few dozen lines of code.

⚡ Rapid reproduction and extensibility

Built on the MCP architecture, all modules are encapsulated as independent and reusable Servers. Each function is registered as a Tool, allowing new functionalities to be seamlessly integrated into the overall pipeline with minimal effort.

📊 Unified evaluation and benchmarking

Provides standardized evaluation workflows and metric management, supports multiple mainstream Benchmark datasets out of the box, and continuously integrates new baselines to enhance research reproducibility and optimization efficiency.