MCP Local RAG

MCP Local RAG is a search server that queries the web through nine different backends—DuckDuckGo, Brave, Searx, Tavily, and others—without needing API keys or cloud dependencies. It runs entirely on your machine, scrapes results, chunks them, and hands back context you can feed into Claude or another LLM. The pitch is simple: fresh web data for research, summarisation, or fact-checking, all offline-friendly and free. I'd reach for this when I'm prototyping something that needs current information but I don't want to burn API credits or leak queries to a third party. It's particularly useful for one-off research tasks—say, comparing product specs, pulling recent news on a niche topic, or checking if a library's documentation has changed. The multi-engine fallback is clever: if DuckDuckGo rate-limits you, it tries Brave, then Searx, and so on. That redundancy matters when you're querying aggressively. The quirks are real. This isn't a polished product. You're running web scrapers locally, which means you inherit all the fragility of HTML parsing. Some backends will time out. Some will return garbage. The chunking logic is basic—don't expect semantic splitting or anything clever. It's also Python 3.10-only, which is fine if you're already in that ecosystem but annoying if you're not. The repo is light on documentation, so you'll need to read the code to understand what each backend actually does. Who shouldn't bother: anyone expecting production-grade reliability or semantic search. This is a developer tool for people who understand the trade-offs of scraping and are comfortable debugging when a backend misbehaves. If you need guaranteed uptime or structured data, pay for a proper search API. But if you're building a prototype, doing ad-hoc research, or just want to avoid vendor lock-in, this is a solid starting point. It does one thing—fetch and chunk web results—and does it without asking for your credit card.