#2525th of 42 in RAG, knowledge and search
ultrarag
MCP-based RAG framework with YAML pipelines and a visual pipeline builder
- Stars
- 5.7k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Last release
- Sep 2026
- Language
- Python
5,345 on 10 Apr 20265,737 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
UltraRAG splits RAG components such as retriever and generation into independent MCP servers, and an MCP client orchestrates them from YAML files with sequence, loop and conditional-branch control. It includes a web UI for building pipelines on a canvas or in code, managing knowledge bases, and turning a pipeline into a chat interface, plus built-in evaluation workflows and benchmark datasets for research.
Who it is for: RAG researchers and developers prototyping retrieval pipelines
Strengths
- Loops and conditional branches in RAG pipelines are defined in YAML, not Python
- Each component is an MCP server, so new tools register at function level
- Built-in evaluation workflows and public benchmark datasets for reproducible RAG experiments
- Pipeline Builder syncs canvas and code editing, and exports a pipeline as a chat UI
Weaknesses
- Aimed at research and prototyping; production hardening is not documented in the README
- Full install pins Linux GPU dependencies to CUDA 12.9 with vLLM
- Docker latest tags are moving nightly builds; pinning a digest is needed for reproducibility
- Some docs, such as the debugging guide, are Chinese-only
What it needs
- GPU optional
- Docker
- Needs uv, Milvus, Docker, NVIDIA Container Toolkit
- Models: vLLM, AgentCPM-Report, MiniCPM-Embedding-Light
- port 5050
Quick answers
Is ultrarag open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Can you run ultrarag with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does ultrarag need a GPU?
No. It runs on a CPU and can use a GPU when you have one.
Also in RAG, knowledge and search
See all 42| Rank | Project | Score |
|---|---|---|
| 1 | RAGFlowRAG engine with deep document parsing, template chunking and agentic retrieval | 90 out of 100 |
| 2 | LightRAGGraph-based RAG framework with a REST API server and web UI | 89 out of 100 |
| 3 | deeptutorSelf-hosted AI tutor with knowledge bases, guided learning, books and quizzes | 88 out of 100 |
| #4 | MaxKBEnterprise knowledge-base agent platform with RAG, workflows and MCP tools | 86 out of 100 |
| #5 | quivrGo engine that ingests content streams for hybrid search and keyword alerts | 84 out of 100 |