mcp-local-rag
Free本地运行的 RAG 式网页搜索，使用 MediaPipe Embedder 和 DuckDuckGo。
FreeFree tier
Inputs: text
About mcp-local-rag
A RAG-based web search and deep research model context protocol (MCP) server that runs entirely locally. Features multi-engine research across 9+ search backends with semantic similarity ranking, and requires no API keys.
Key Features
Multi-Engine Deep Research with 9+ search backends (DuckDuckGo, Google, Bing, Brave, Wikipedia, Yahoo, Yandex, Mojeek, Grokipedia)
Semantic similarity ranking using Google's MediaPipe Text Embedder
No API keys required; runs entirely locally with embedded models
Privacy options: choose privacy-focused engines (DuckDuckGo, Brave) or comprehensive ones (Google)
Tools: deep_research, deep_research_google, deep_research_ddgs, rag_search_ddgs, rag_search_google
Pros & Cons
Pros
- No API keys or external services required; fully local
- Privacy-friendly search options (DuckDuckGo, Brave)
- Semantic relevance ranking improves result quality
- Supports 9+ search backends for broad coverage
- Runs entirely locally with embedded models and no internet dependency for ranking
Cons
- Primitive RAG-like implementation may not match full RAG systems in sophistication
- Requires uv or Docker for setup and execution
- Search results depend on free search engine availability and rate limits
Best For
Comprehensive multi-engine research for complex topics requiring diverse perspectivesPrivacy-focused web search without trackingTechnical and scientific queries via Google-focused deep diveQuick single searches with rag_search tools for fast answersDeep research with customizable backends and result limits
FAQ
Does it require API keys?
No, all processing runs locally with no API keys needed. The server uses embedded models and free search engines.
What search engines are supported?
DuckDuckGo, Google, Bing, Brave, Wikipedia, Yahoo, Yandex, Mojeek, and Grokipedia.
How does the ranking work?
It uses Google's MediaPipe Text Embedder to compute semantic similarity between the query and search results, then ranks entries to select the top k most relevant.
Can I use it with Claude Desktop?
Yes, examples for Claude Desktop are provided in the repository.