mcpsurfr
Provide local web search, content extraction, and reranking for language models without requiring API keys.
mcpsurfr (formerly mcp-local-rag) is a privacy-first web search and retrieval tool designed to enhance LLM capabilities locally. Operating without third-party API keys, it queries multiple search engines via ddgs, extracts clean webpage content (including PDFs and documentation from llms.txt), and reranks the results directly on your machine using an embedded MobileBERT model and BM25. It bypasses scraper blocks on popular developer platforms through dedicated free APIs and optimizes context consumption by returning precise passages instead of bloated full pages.
主な機能
01On-device reranking using BM25 and a local MobileBERT model running on NumPy
02Smart web scraping with browser impersonation, boiler-plate removal, PDF reading, and caching
03Direct API integrations for scraper-resistant sites like GitHub, Stack Overflow, and Hacker News
04Passage-level extraction to reduce context token usage by returning concise relevant paragraphs
05Keyless multi-engine web search querying DuckDuckGo, Brave, Google, and more via ddgs
06151 GitHub stars
ユースケース
01Supplying local LLMs with real-time web knowledge and breaking news without commercial search API fees
02Conducting deep automated research across multiple terms and backends with de-duplicated results
03Extracting specific code snippets, package info, and technical answers from developer platforms directly into LLM context